Storage system

The storage system optimizes data transfer by using compression and decompression circuits to manage data transfer efficiently, addressing performance bottlenecks and reducing response time in multi-node storage systems.

JP2024165412AInactive Publication Date: 2024-11-28HITACHI VANTARA LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
JP2023081593
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-17
Publication Date
2024-11-28
Estimated Expiration
Not applicable · inactive patent

AI Technical Summary

Technical Problem

Conventional storage systems face performance deterioration due to the time required for compression/decompression processing, which increases the response time and reduces the communication bandwidth between storage nodes, leading to bottlenecks in data transfer.

Method used

A storage system with multiple storage nodes equipped with processors, drives, and communicators that include compression and decompression circuits to reversibly compress and decompress data during transfer, optimizing data transfer based on conditions such as data size and CPU utilization to avoid unnecessary processing time.

Benefits of technology

The solution effectively suppresses performance deterioration by dynamically managing compression and decompression, improving communication bandwidth and reducing response time, thereby enhancing the overall system performance.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024165412000001_ABST
    Figure 2024165412000001_ABST
Patent Text Reader

Abstract

To suppress deterioration of response performance of a storage system.SOLUTION: Each storage node of multiple storage nodes includes: a processor for processing instructions from the outside; a drive for storing data; and a communication device for transmitting data to another storage node or receiving data from another storage node. The communication device includes a compression circuit for reversibly compressing data before transmitting them and an extension circuit for extending the compressed data after receiving them. According to a read instruction for reading data of a designated size to the outside, the communication device of a first storage node compresses data stored in the drive of the first storage node by the compression circuit and transmits them to the communication device of a second storage node when a predetermined condition is satisfied. The communication device of the second storage node extends the received data by the extension circuit. The second storage node outputs the extended data to the outside.SELECTED DRAWING: Figure 8
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to a storage system. [Background technology]

[0002] In a storage system, which is an information device for storing and managing large amounts of data, it is preferable to be able to easily expand the data capacity when the need arises to store more data.Therefore, there are storage systems that are designed to be able to interconnect multiple storage nodes, so that the number of connected storage nodes can be increased later as needed.This is called a multi-node connection configuration storage system.

[0003] In such a storage system, a host that instructs data reading and writing is connected to each storage node. When data is read by a read command from the host, if that data is stored in a storage node to which the host is not connected, data transfer between storage nodes is required. When data is written by a write command from the host, data transfer must be performed to back up the data to another storage node in case the storage node to which the host is connected fails.

[0004] In a multi-node connected storage system, if the communication bandwidth between storage nodes is not large enough, the bandwidth will become a bottleneck in the host's data read and write performance. Therefore, it is desirable to expand the communication bandwidth between storage nodes. There are two ways to achieve this: one is to increase the number of communication devices that communicate between storage nodes, and the other is to compress and reduce the amount of communication data between storage nodes.

[0005] The disadvantage of the former method is that the inclusion of a communications device increases the system's production costs. The disadvantage of the latter method is that the time required for command processing increases by the amount of compression and decompression processing, resulting in poor response performance. If the time required for compression and decompression processing can be reduced, it is preferable to choose the latter method, as it is less expensive.

[0006] Examples of connection methods between storage nodes include Ethernet and PCI Express. The communication method used there is to divide data into multiple pieces and embed them in the payload of IP packets or TLP (Transaction Layer Packet) to send and receive them.

[0007] As a conventional technique, Non-Patent Document 1 specifies a protocol for compressing payload data to reduce the amount of IP packets transmitted over Ethernet. This technique has the following features: Each payload is compressed independently using a dictionary compression algorithm. The condition for applying compression is specified only when "the payload size does not increase due to compression." It does not specify how the payload size is determined. [Prior art documents] [Non-patent literature]

[0008] [Non-Patent Document 1] RFC-3173(IP Payload Compression Protocol) Summary of the Invention [Problem to be solved by the invention]

[0009] Conventional technologies related to communication data compression, including Non-Patent Document 1, do not disclose a method for reducing the time required for compression / decompression processing to suppress deterioration in the response performance of a storage system when compressing and reducing the amount of communication data between storage nodes in a storage system. [Means for solving the problem]

[0010] One aspect of the present invention is a storage system including a plurality of storage nodes, each storage node of the plurality of storage nodes including a processor for processing commands from an external source, a drive for storing data, and a communication device for transmitting data to other storage nodes or receiving data from the other storage nodes, the communication device including a compression circuit for losslessly compressing data before transmitting it, and an decompression circuit for decompressing the compressed data after receiving it, and in response to a read command for reading data of a specified size to the outside, if a predetermined condition is satisfied, the communication device of a first storage node compresses data stored in the drive of the first storage node using the compression circuit and transmits it to the communication device of a second storage node, the communication device of the second storage node decompresses the received data using the decompression circuit, and the second storage node outputs the decompressed data to the outside.

[0011] One aspect of the present invention is a storage system including a plurality of storage nodes, each storage node of the plurality of storage nodes including a processor for processing commands from an external source, a drive for storing data, a cache memory, and a communication device for transmitting data to other storage nodes or receiving data from the other storage nodes, the communication device including a compression circuit for losslessly compressing data before transmission and an decompression circuit for decompressing the compressed data after receiving it, and in response to a write command for writing received data of a specified size from the outside, if a specified condition is satisfied, the communication device of a first storage node compresses data in the cache memory of the first storage node using the compression circuit and transmits it to the communication device of a second storage node, the communication device of the second storage node decompresses the received data using the decompression circuit, and the second storage node stores the decompressed data in the cache memory of the second storage node.

[0012] One aspect of the present invention is a storage system including a plurality of storage nodes, each storage node of the plurality of storage nodes including a cache memory and a communication device for transmitting data to other storage nodes or receiving data from the other storage nodes, the communication device including a compression circuit for losslessly compressing data before transmission and an decompression circuit for decompressing the compressed data after receiving it, the communication device of a first storage node divides data into a plurality of partial data using a maximum payload size of a write packet from the communication device to the cache memory as a division unit, and transmits a packet having the partial data compressed by the compression circuit as a payload to the communication device of a second storage node, the communication device of the second storage node receiving the packet decompresses the payload of the received packet using the decompression circuit, constructs a write packet including the decompressed payload, and transfers the write packet to the cache memory of the second storage node. Effect of the Invention

[0013] According to one aspect of the present invention, in a storage system having a compression function for communication data between storage nodes, it is possible to suppress deterioration of response performance. [Brief description of the drawings]

[0014] [Figure 1] 1 shows the configuration of a storage system to which the present invention is applied. [Diagram 2] The configuration of an inter-node communicator is shown. [Diagram 3] 13 shows a flowchart of a read command for data stored in an uncompressed state. [Figure 4] 13 shows the first half of a flowchart of a write command for data to be stored in an uncompressed state. [Diagram 5] 13 shows the second half of the flowchart for a write command for data to be stored in an uncompressed state. [Figure 6] 13 shows a flowchart of a read command for data stored in a compressed state. [Figure 7A]The method for calculating the maximum read performance is shown below. [Figure 7B] This shows the change in read performance relative to CPU utilization. [Figure 8] 13 shows a flowchart for determining a compression condition for inter-node transfer. [Figure 9A] The figure shows the data flow of compression and decompression using the inter-node communication device. [Figure 9B] An example of dictionary compression is shown below. [Figure 10] This shows division and compression of data transferred between nodes. [Figure 11] 1 shows the configuration of a packet transmitted and received by an inter-node communicator. [Figure 12] 1 shows a flowchart of a process performed by a compression circuit. [Figure 13] 13 shows a flowchart of the process performed by an inter-node communicator on a received packet. [Figure 14] 13 shows a change in transmission output rate from an inter-node communicator in the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, the embodiments will be described with reference to the drawings. Note that the embodiments are merely examples for realizing the present invention, and do not limit the technical scope of the present invention, and all of the combinations of features described in the embodiments are not necessarily essential to the solution of the invention.

[0016] In the following description, various types of information may be described using the expression "xxx table", but the various types of information may be expressed in a data structure other than a table. To indicate that it does not depend on the data structure, the "xxx table" may be called "xxx information". Also, in the following description, numbers are used as identification information for elements, but other types of identification information (e.g., names, identifiers) may be used.

[0017] In addition, in the following description, when describing elements of the same type without distinguishing between them, common symbols (or reference signs) may be used, and when describing elements of the same type with distinction between them, reference signs (or element IDs) may be used.

[0018] The program is executed by a processor (e.g., a CPU (Central Processing Unit)) included in the storage controller to perform a defined process using a storage resource (e.g., a main memory) and / or a communication interface device as appropriate, so the subject of the process may be the storage controller or the processor. The storage controller may also include a hardware circuit that performs some or all of the process. The computer program may be installed from a program source. The program source may be, for example, a program distribution server or a computer-readable storage medium.

[0019] (1) System Configuration As an embodiment of this specification, a storage system with a multi-node connection configuration will be described. Fig. 1 shows the internal configuration of a storage system 100 with a multi-node connection configuration. In this system, a plurality of storage nodes 101A, 101B are connected via a hub device 110 so as to be capable of data communication with each other, and hosts 108A, 108B that request reading and writing of data to be stored in the storage system 100 by read / write commands are connected to the storage nodes 101A, 101B, respectively.

[0020] The number of storage nodes is not limited to two as shown in Fig. 1, but may be two or more. For example, if the number is four, four storage nodes are connected to the hub device 110, and data communication is possible between any two of the storage nodes. The communication device between the storage nodes is not limited to the hub device 110, and other communication devices may be used in addition to or instead of the hub device 110.

[0021] In the embodiment described below, the internal configuration of each storage node is the same, and the details of the components will be described below by abbreviating the A or B at the end of the number.

[0022] The storage node 101 includes a host I / F (Interface) 102 , a CPU 103 which is a processor, an inter-node communicator 104 , a data storage medium 105 , a cache memory 106 , and a stored data compressor / decompressor 107 .

[0023] The host I / F 102 is an interface mechanism for connecting to a host 108, and responds to read / write commands from the host in order to transmit data to the host and receive data from the host. The mechanism of the host I / F 102 and the protocol for transmitting and receiving commands and data comply with a standard interface standard, for example, the FibreChannel standard.

[0024] The data storage medium 105 is, for example, a hard disk drive (HDD) or a solid state drive (SSD) equipped with a NAND flash memory, which is a non-volatile semiconductor memory, has a large capacity, and permanently stores data received from the host. The data storage medium 105 is also called a storage drive or simply a drive.

[0025] The cache memory 106 uses a volatile memory medium such as a DRAM (Dynamic Random Access Memory), and temporarily stores data received from the host 108 or data read from the data storage medium 105.

[0026] The stored data compressor / decompressor 107 reversibly compresses write data received in response to a write command to generate compressed data in order to reduce the amount of data stored in the data storage medium 105. Also, in order to transmit the original plain data to the host 108 in response to a read command, the stored data compressor / decompressor 107 decompresses compressed data read from the data storage medium 105 to generate plain data. Whether or not compression by the stored data compressor / decompressor 107 is applied to data stored in the data storage medium 105 can be changed by the initial setting of the storage node 101.

[0027] The CPU 103 is connected to the host I / F 102, the inter-node communicator 104, the data storage medium 105, the cache memory 106, and the stored data compressor / decompressor 107, and includes a plurality of microprocessors that control them. The CPU 103 executes data transfer between the inter-node communicator 104, the data storage medium 105, and the cache memory 106.

[0028] The data transfer protocol complies with a standard interface standard, for example, the PCI Express standard. In this case, the CPU 103 functions as a RootComplex (parent), and the inter-node communicator 104 and the data storage medium 105 function as EndPoints (children). In the PCI Express standard, data to be transferred is divided into a predetermined size and embedded in the payload portion of a transfer format called a packet before being transferred. The maximum payload size supported by the CPU 103 is, for example, 512 bytes.

[0029] The CPU 103 interprets the contents of a read / write command from the host 108. It also instructs the stored data compressor / decompressor 107 to compress and decompress data. It also instructs data transfer between the inter-node communicator 104 or data storage medium 105 and the cache memory 106.

[0030] In each storage node 101, write data from the host 108 is first temporarily stored in the cache memory 106. If the data is initially set to be stored in the data storage medium 105 in an uncompressed state, the data is written to the data storage medium 105 as is. On the other hand, if the data is initially set to be stored in the data storage medium 105 in a compressed state, the data is converted into compressed data by the stored data compressor / decompressor 107, and the compressed data is temporarily stored in the cache memory 106. The compressed data is then written to the data storage medium 105.

[0031] In each storage node, read data to host 108 is read from data storage medium 105 in an uncompressed or compressed state, and first temporarily stored in cache memory 106. If the data is in an uncompressed state, it is sent to host 108 as is. On the other hand, if the data is in a compressed state, it is converted to plain data by stored data compressor / decompressor 107, and the plain data is temporarily stored in cache memory 106. The plain data is then sent to host 108.

[0032] The inter-node communicator 104 is used when data transfer is required between storage nodes while performing processing in response to a read / write command from the host 108. For example, the host 108A requests the storage node 101A to read data stored in the data storage medium 105B in the storage node 101B other than the storage node 101A to which the host 108A is connected. The data is transferred from the storage node 101B to the storage node 101A to which the host 108A is connected. Details including other cases will be described later.

[0033] The protocol for data transfer between storage nodes by the inter-node communicator 104 complies with a standard interface standard, for example, the PCI express standard. In this case, the hub device 110 functions as a RootComplex (parent), and the inter-node communicator 104 functions as an EndPoint (child). (2) Inter-node communication device

[0034] The internal configuration of the inter-node communicator 104 will be described. Fig. 2 shows the internal configuration of the inter-node communicator 104. The inter-node communicator 104 includes a PCIexpress compliant interface (hereinafter, PCIe I / F) 201 that connects to the CPU 103, and a PCIe I / F 202 that connects to the hub device 110. For example, when transferring data from the storage node 101A to 101B, the inter-node communicator 104A inputs data from the CPU 103A via the PCIe I / F 201, and transmits the data from the PCIe I / F 202 to the inter-node communicator 104B via the hub device 110. Then, the inter-node communicator 104B receives the data via the PCIe I / F 202, and outputs the data from the PCIe I / F 201 to the CPU 103B.

[0035] The inter-node communicator 104 includes four types of buffers 210, 213, 220, and 223 for relaying transfer data between the PCIe I / F 201 and the PCIe I / F 202. The transmission input buffer 210 is a buffer memory for receiving data from the CPU 103. The transmission output buffer 213 is a buffer memory for storing data to be transmitted to other storage nodes. The reception input buffer 220 is a buffer memory for holding data received from other storage nodes. The reception output buffer 223 is a buffer memory for storing data to be transmitted to the CPU 103.

[0036] When transferring data from the transmission input buffer 210 to the transmission output buffer 213, the inter-node communication device 104 can transfer the data without processing it or transfer it after lossless compression, according to an instruction from the CPU 103. For the latter lossless compression, a plurality of compression circuits 212 are provided. When transferring data after lossless compression, the plurality of compression circuits 212 execute compression processing in parallel. The compression buffer 211 is a buffer memory provided for each compression circuit 212.

[0037] The transfer data is divided into pieces of a predetermined size and distributed to each compression circuit 212. The compression buffer 211 temporarily holds these divided data. The multiple compression results obtained from the compression circuit 212 are transferred to the transmission output buffer 213 and combined again. On the other hand, when transferring data without processing, the data is transferred to the transmission output buffer 213 without passing through the compression buffer 211 or compression circuit 212.

[0038] When transferring data from the receiving input buffer 220 to the receiving output buffer 223, the inter-node communication device 104 can transfer the data without processing it or can decompress the data before transferring it, depending on whether the data was compressed by the inter-node communication device 104 that is the transmission source. For the latter decompression, a plurality of decompression circuits 222 are provided. When decompressing and transferring the data, the plurality of decompression circuits 222 execute the decompression process in parallel. The decompression buffer 221 is a buffer memory provided for each decompression circuit 222.

[0039] The transfer data is divided and distributed to each decompression circuit 222. The decompression buffer 221 temporarily holds the divided data. The multiple decompression results obtained from the decompression circuit 222 are all of the specified size when divided before lossless compression, and are transferred to the reception output buffer 223 and reassembled to return to the size before lossless compression. On the other hand, when transferring without processing, the data is transferred to the reception output buffer 223 without passing through the decompression buffer 221 or the decompression circuit 222.

[0040] For example, if the inter-node communicator 104 is equipped with 32 compression circuits 212 and 256 KB of data is losslessly compressed and transferred to another storage node, the data is divided into 8 KB, and each compression circuit 212 performs lossless compression on 8 KB. If the compression rate of the lossless compression by the compression circuit 212 is an average of 50%, the size of each compression result will be 4 KB on average. The size of the compressed data collected in the transmission output buffer 213 will be 128 KB. This compressed data is sent to the inter-node communicator 104 of the other storage node.

[0041] The inter-node communicator 104 on the receiving side decompresses the received data in parallel using multiple decompression circuits 222 to restore the original data, and transfers it to the CPU 103. At this time, the size of each decompression result by the decompression circuits 222 is 8 KB. Then, 32 pieces are collected in the reception output buffer 223, and the original 256 KB data is obtained.

[0042] The inter-node communicator 104 can reduce the size of data transferred between storage nodes by using the provided compression circuit 212 and decompression circuit 222. When the compression rate of the lossless compression by the compression circuit 212 is 50%, this means that the transfer bandwidth between storage nodes is apparently doubled. When the performance bottleneck of the storage system 100 is the transfer bandwidth between storage nodes, applying this compression improves the bottleneck and brings about an improvement in system performance.

[0043] (3) Read / Write Commands The read / write command processing performed by the storage node 101 will be described with reference to FIGS.

[0044] FIG. 3 shows a flowchart of a process in which the storage node 101A reads data stored in an uncompressed state in the data storage medium 105B in the storage node 101B in response to a read command from the host 108A and returns the data to the host 108A.

[0045] First, storage node 101A receives a data read command from host 108A (301). Since the requested data is in storage node 101B, storage node 101A requests storage node 101B to read the data via inter-node communicators 104A and 104B (302).

[0046] Next, storage node 101B reads the requested data from data storage medium 105B (303) and stores the data in cache memory 106B (304). CPU 103B then determines whether the condition for compressing the data during inter-node communication is met (305). Details of this condition will be described later. If the condition is not met (NO), CPU 103B instructs inter-node communicator 104B to transfer the data to storage node 101A uncompressed (unprocessed) (307), and inter-node communicator 104B transmits the data to storage node 101A as is (309).

[0047] If the condition is met in step 305 (YES), the CPU 103B instructs the inter-node communicator 104B to transfer compressed data to the storage node 101A (306), and the inter-node communicator 104B compresses the data using the compression circuit 212 in FIG. 2 (308) and then transmits the data (309).

[0048] Next, the inter-node communicator 104A of the storage node 101A receives the data (310). The inter-node communicator 104A determines whether the received data has been compressed by the inter-node communicator 104B (311). The transferred data includes data indicating whether it has been compressed or not. If it has not been compressed (unprocessed), it is stored as is in the cache memory 106A (313), but if it has been compressed, it is decompressed by the decompression circuit 222 in FIG. 2 (312) and then stored in the cache memory 106A (313). Finally, the storage node 101A returns the data to the host 108A as a response to the read command (314).

[0049] 4 and 5 show a flowchart of a process in which the storage node 101A writes data to the data storage medium 105A in the storage node 101A in response to a write command from the host 108A.

[0050] 4, storage node 101A receives a data write command and data to be written from host 108A (401), and stores the data in cache memory 106A (402). To prevent loss of the write data due to a failure of storage node 101A (excluding data storage medium 105A), this write data is also stored as a backup in cache memory 106B in storage node 101B. The following steps 403 to 412 are the procedure for transferring the write data to storage node 101B via inter-node communicators 104A and 104B, and storing it in cache memory 106B.

[0051] CPU 103A determines whether the condition for compressing the data during inter-node communication is met (403). Details of this condition will be described later. If the condition is not met (NO), CPU 103A instructs inter-node communicator 104A to transfer the data to storage node 101B without compression (no processing) (405), and inter-node communicator 104A transmits the data to storage node 101B as is (407).

[0052] If the condition is met in step 403 (YES), the CPU 103A instructs the inter-node communicator 104A to transfer compressed data to the storage node 101B (404), and the inter-node communicator 104A compresses the data using the compression circuit 212 in FIG. 2 (406) and then transmits the data (407).

[0053] Next, the inter-node communicator 104B of the storage node 101B receives the data (408). The inter-node communicator 104B determines whether the received data has been compressed by the inter-node communicator 104A (409). The transferred data includes data indicating whether it has been compressed or not, and this data is referenced. If the data has not been compressed (unprocessed), it is stored as is in the cache memory 106B (411), but if it has been compressed, it is decompressed by the decompression circuit 222 in FIG. 2 (410) and then stored in the cache memory 106B (411). The storage node 101B notifies the storage node 101A via the inter-node communicators 104B and 104A that it has completed storing the write data (412).

[0054] Next, storage node 101A transmits a data write completion response to host 108A (413). Although data storage in data storage medium 105A is not actually complete, the response is sent at this point so that host 108A can prepare the next command early.

[0055] 5, the CPU 103A of the storage node 101A calculates RAID parity from the write data (501). The write data to be stored in the data storage medium 105A is recorded after being made redundant based on RAID (Redundant Arrays of Independent Disks) technology in order to prevent data loss due to a medium failure.

[0056] Specifically, if there are N data storage media 105A, CPU 103A will distribute the write data evenly across (N-1) media, and record the parity created by calculating the exclusive OR of this data in the remaining media. This makes it possible to recover data even if one of the N media fails.

[0057] For example, when N=4, CPU 103A records data of the same size D1, D2, and D3 on three drives, and records RAID parity P calculated by P=D1+D2+D3 (+ indicates exclusive OR) on the remaining drive. For example, if the recording medium of D2 fails, CPU 103A recovers D2 by utilizing the property that P+D1+D3=D2. Since parity P is created by the exclusive OR of different data, its contents are generally meaningless and random. In step 502, CPU 103A stores the calculated parity in cache memory 106A.

[0058] To prevent loss of parity due to a failure of storage node 101A (excluding data storage medium 105A), this parity is also held as a backup in cache memory 106B in storage node 101B. The following steps 503 to 508 are the procedure for transferring the parity to storage node 101B via inter-node communicators 104A and 104B and holding it in cache memory 106B.

[0059] The CPU 103A instructs the inter-node communicator 104A to transfer the parity to the storage node 101B without compression (without using the compression circuit 212 in FIG. 2) (503). The decision to make this transfer uncompressed will be described later. The inter-node communicator 104A sends the parity as is to the storage node 101B (504).

[0060] Next, the inter-node communicator 104B of storage node 101B receives the parity (505). The inter-node communicator 104B determines whether the parity has been compressed by the inter-node communicator 104A (506). The transferred data includes data indicating whether it has been compressed or not, and this data is referenced. In this case, the data has not been compressed by the inter-node communicator 104A, so it is stored as is in the cache memory 106B (507). The storage node 101B notifies storage node 101A via the inter-node communicators 104B and 104A that it has completed storing the parity (508).

[0061] Finally, the storage node 101A stores the write data and parity held in the cache memory 106A in the data storage medium 105A (509). Since there is no possibility of the write data being lost after this, the backup held in the cache memory 106B is invalidated.

[0062] FIG. 6 shows a flowchart of a process in which, in response to a read command from host 108A, storage node 101A reads data stored in a compressed state by data compressor / decompressor 107B in data storage medium 105B in storage node 101B, decompresses the data to restore it to a plain text state, and returns the data to host 108A.

[0063] First, the storage node 101A receives a data read command from the host 108A (601). Since the requested data is in the storage node 101B, the storage node 101A requests the storage node 101B to read the data via the inter-node communicators 104A and 104B (602).

[0064] Next, storage node 101B reads the requested compressed data from data storage medium 105B (603) and stores the compressed data in cache memory 106B (604). CPU 103B then instructs inter-node communicator 104B to transfer data to storage node 101A without compression (compression circuit 212 in FIG. 2 is not used) (605). The decision to make this transfer uncompressed will be described later. Inter-node communicator 104B transmits the compressed data as is to storage node 101A (606).

[0065] Next, the inter-node communicator 104A of the storage node 101A receives the compressed data (607). The inter-node communicator 104A determines whether the data has been compressed by the inter-node communicator 104B (608). The transferred data includes data indicating whether it has been compressed or not, and this data is referenced. In this case, the data has not been compressed by the inter-node communicator 104B, so the data is stored as is in the cache memory 106A (609). The storage node 101A decompresses the compressed data using the stored data compressor / decompressor 107A, restoring it to a plain text state and storing it in the cache memory 106A (610). Finally, the data is returned to the host 108A as a response to the read command (611).

[0066] (4) Effect and control of compression by inter-node communication device The effect of compressing communications between storage nodes and control of enabling / disabling the compression will be described with reference to FIGS. 7A, 7B, and 8. FIG.

[0067] 7A is a table showing a method for calculating the maximum performance (maximum read performance, in GB / s) when the storage node 101 returns data in response to a read command received from the host 108. The maximum read performance differs depending on the requested data size d (kB) per read command. The time T (s) it takes one microprocessor in the CPU 103 to process one read command is, for example, T=80 when d=8, T=200 when d=32, and T=800 when d=256.

[0068] If the number of cores c of the microprocessor included in CPU 103 is 40, the read command processing performance of CPU 103 is maximum when all cores are operating, and the performance (maximum CPU performance, in GB / s) is calculated by dc / T. For example, it is 4.0 GB / s when d=8, 6.4 GB / s when d=32, and 12.8 GB / s when d=256. Even if CPU 103 has such processing performance, if the transfer bandwidth of the path through which read data is sent to host 108 is less than the maximum CPU performance, the transfer bandwidth will become a bottleneck and the maximum read performance will be lower than the maximum CPU performance.

[0069] Now, if all read commands from the host 108 request data stored in the data storage medium 105 in another storage node 101 to which the host 108 is not connected, all of that data is transferred via the inter-node communicator 104. Suppose that this transfer bandwidth (inter-node bandwidth) is the lowest among the paths through which the read data passes, for example, 5.0 GB / s. The maximum read performance in this case is calculated. When d=8, the maximum CPU performance is 4.0 GB / s, so the inter-node bandwidth does not become a bottleneck, and it becomes 4.0 GB / s. When d=32 or 256, the maximum CPU performance is 6.4 GB / s or 12.8 GB / s, so the inter-node bandwidth becomes a bottleneck, and both are suppressed to 5.0 GB / s.

[0070] In contrast, we calculate the maximum read performance when the inter-node communicator 104 losslessly compresses the transferred data and doubles the apparent bandwidth to 10.0 GB / s when d=32 or 256. When d=32, the inter-node bandwidth is not a bottleneck and the maximum read performance improves to 6.4 GB / s. When d=256, the inter-node bandwidth is still a bottleneck, but the maximum read performance improves to 10.0 GB / s.

[0071] Fig. 7B shows the change in read performance of the storage node 101 relative to the operating rate of the CPU 103 (number of operating cores / total number of cores). A control method in which the CPU 103 enables / disables compression by the inter-node communicator 104 will be described with reference to Fig. 7B. The CPU operating rate referred to is the operating rate of the CPU 103 of the storage node 101 that received a read command or write command from the host 108. The number of operating cores can be controlled from the standpoint of power consumption and other factors. This is a widely known technique and details will be omitted.

[0072] First, when d=8, the inter-node bandwidth does not become a bottleneck in read performance even if the CPU operating rate increases, so the CPU 103 instructs the inter-node communicator 104 to transfer data without compression. The read performance can be improved up to the maximum CPU performance of 4.0 GB / s.

[0073] When d=32, if the CPU utilization rate is 78% (border line 711) or less, the internode bandwidth will not become a bottleneck in read performance, so the CPU 103 instructs the internode communicator 104 to transfer data without compression, but if the CPU utilization rate is greater than 78% (border line 711), the CPU 103 instructs the internode communicator 104 to transfer data with compression in order to prevent the internode bandwidth from becoming a bottleneck in read performance. Compressing internode communications can improve read performance up to the maximum CPU performance of 6.4 GB / s.

[0074] When d=256, if the CPU utilization rate is 39% (border line 712) or less, the internode bandwidth will not become a bottleneck in read performance, so the CPU 103 instructs the internode communicator 104 to transfer data uncompressed, but if the CPU utilization rate is greater than 39% (border line 712), the CPU 103 instructs the internode communicator 104 to transfer data compressed to prevent the internode bandwidth from becoming a bottleneck in read performance. Compressing internode communications makes it possible to improve read performance up to the internode bandwidth (apparent bandwidth) of 10.0 GB / s.

[0075] As described above, by setting different CPU utilization rate thresholds according to the requested data size d, more appropriate judgment is possible. A common CPU utilization rate threshold may be set for different requested data sizes d. With the above control method, when the compression / decompression processing by the inter-node communicator 104 is not effective in improving read performance, the time required for these processing can be saved, and unnecessary increases in the response time until the read data is returned to the host 108 can be suppressed.

[0076] The explanation with reference to Figures 7A and 7B may be applied to the case where the storage node 101 returns a data write completion response to a write command received from the host 108. As explained with reference to Figure 4, data received from the host 108 is transferred to another storage node 101 via the inter-node communicator 104. Whether or not to compress the data in the inter-node communicator 104 is determined so that the inter-node bandwidth does not become a bottleneck in the write performance. The CPU 103 of the storage node 101 that has received the write command from the host 108 determines whether or not to compress the data in the inter-node communicator 104 based on the relationship between the operation rate and a specified threshold value. As described above, one or more threshold values ​​are set.

[0077] FIG. 8 shows a flowchart of steps (ie, 305, 403, 503, 605) in the read / write command processing shown in FIGS. 3 to 6, in which the CPU 103 determines the conditions (compression conditions) for enabling compression of inter-node communications.

[0078] CPU 103 checks whether the transfer target is RAID parity (801), and if true, ends the determination by declaring that the compression conditions are not met (806). If step 801 is false, CPU 103 checks whether the transfer target is already compressed data (802), and if true, ends the determination by declaring that the compression conditions are not met (806). If step 802 is false, CPU 103 checks whether the size of the data requested to be read / written by the command is equal to or larger than a threshold (803). In this example, the condition is 32 kB or more. If step 803 is false, ends the determination by declaring that the compression conditions are not met (806).

[0079] If step 803 is true, it is checked whether the CPU operating rate of the storage node 101 that received the read / write command from the host 108 is greater than a predetermined threshold (804). If it is false, the compression conditions are not met (806) and the determination is terminated. If 804 is true, the compression conditions are met (805) and the determination is terminated.

[0080] The judgments in steps 801 and 802 are based on the fact that RAID parity and compressed data are meaningless and random, and have a low compression ratio, so that spending time on compression and decompression has no effect on expanding the internode bandwidth and only increases the response time. The judgment in step 803 is based on the fact that when the data size is small, the internode bandwidth does not become a bottleneck in read / write performance, and therefore it is not necessary to expand the internode bandwidth. The judgment in step 804 is based on the fact that when the CPU operating rate is equal to or less than a threshold value (711 or 712 in FIG. 7B), the internode bandwidth does not become a bottleneck in read / write performance, and therefore it is not necessary to expand the internode bandwidth. Note that only some of the conditions shown in FIG. 8 may be judged, and all of the conditions may not be judged.

[0081] (5) Compression and decompression method using inter-node communication device 9A shows an overview of a data compression / decompression method by the inter-node communicator 104. In the compression process performed by the inter-node communicator 104 in the source storage node 101, uncompressed plain text data 901 is first subjected to dictionary compression process 902. The dictionary compression result is then subjected to encoding process 903 into a bit string. This generates compressed data 904 that is output from the inter-node communicator 104 to the destination storage node 101.

[0082] On the other hand, in the decompression process performed by the inter-node communicator 104 in the destination storage node 101, the compressed data 904 is first subjected to a bit string decryption process 905. Then, the decrypted result is subjected to a plaintext expansion process 906. As a result, the original plaintext data 901 is generated.

[0083] 9B shows a specific example of dictionary compression processing 902. In the string stream of plaintext data 901, it is checked in order whether the same string appears again. If a certain string matches the first character of the string for L consecutive characters starting from J characters before the first character, this string is converted to a copy symbol [J,L].

[0084] For example, in a five-character string 911 of "a,b,c,d,e," there is a match for five consecutive characters starting from six characters before the first character "a." In this case, the string 911 is converted to a copy symbol [6,5]. Similarly, in a four-character string 912 of "a,b,a,b," there is a match for four consecutive characters starting from two characters before the first character "a" (including overlapping parts). In this case, the string 912 is converted to a copy symbol [2,4]. Similarly, in a four-character string 913 of "c,d,e,f," there is a match for five consecutive characters starting from 15 characters before the first character "c." In this case, the string 913 is converted to a copy symbol [15,5].

[0085] The amount of data for these copy symbols is less than the amount of data for the original string, so this conversion can reduce the amount of data. The range of the string stream (hereafter referred to as the dictionary) referenced in the match search ranges from one character before to a specified number of characters before. Since the dictionary range slides backwards with each search, this compression technique is also called sliding dictionary compression. Note that if there are multiple matching strings within the dictionary range, the longest consecutive matching string is converted to a copy symbol. This has the effect of further reducing the amount of data.

[0086] In the latter encoding process 903, characters that have not been converted to copy symbols (hereafter referred to as literal characters) and copy symbols are encoded with a prescribed bit pattern, and then concatenated to generate a bit stream. The bit stream in Fig. 9B is the result of encoding according to the specifications of the LZS (Lempel-Ziv-Stac) compression algorithm defined in RFC-2395. The copy symbol is encoded by concatenating a bit pattern representing the distance J to the copy source and the copy length L after one bit of "1".

[0087] For example, bit pattern 921 is 13 bits long and represents the copy symbol [6,5]. Bit pattern 922 is 11 bits long and represents the copy symbol [2,4]. Bit pattern 923 is 13 bits long and represents the copy symbol [15,5]. The bit length of the code corresponding to the copy symbol is not fixed. On the other hand, a literal character is represented by a 9-bit bit pattern in which one bit "0" is added to the beginning of the 8-bit value of the character.

[0088] The decryption process 905 interprets such a bit stream and outputs copy symbols and literal characters. Furthermore, the plaintext expansion process 906 restores the character string of the plaintext data 901 from the beginning, in order, from the copy symbols and literal characters. When expanding copy symbols into character strings, the restored character string is referenced as a dictionary. When expanding a copy symbol [J,L], L character is extracted from a position J characters back from the end of the dictionary, and the dictionary with this L character added to the end of the dictionary is used as the dictionary to be referenced next. When expanding a literal character, that one character is added to the end of the dictionary and used as the dictionary to be referenced next.

[0089] Using FIG. 10, we will explain a method for minimizing the processing time (hereinafter referred to as compression overhead time) that increases when the inter-node communicator 104 compresses and transfers inter-node transfer data compared to when the data is transferred uncompressed.

[0090] As described with reference to FIG. 2, the inter-node communicator 104 divides the inter-node transfer data into predetermined sizes and compresses them in parallel by a plurality of compression circuits 212. If the size of the inter-node transfer data 1000 is 256 kB and the predetermined size is 8 kB, the inter-node transfer data 1000 is divided into 32 8 kB portions 1001. Each 8 kB portion 1001 is losslessly compressed by one compression circuit 212. If the inter-node communicator 104 is equipped with 32 compression circuits 212, 32 parallel compression processes are possible, so the time required to compress the 256 kB transfer data 1000 is the same as the time required to compress one 8 kB portion 1001. In other words, by providing a plurality of compression circuits 212 and operating them in parallel, the compression overhead time can be reduced.

[0091] Also, the inter-node communicator 104 divides the 8 kB portion 1001 into 16 512 B portions 1002, and losslessly compresses each of the 512 B portions 1002 individually. For example, 1003 in FIG. 10 shows this. As soon as the compression of each 512 B portion 1002 is completed, a PCI Express packet is formed with the compression result as the payload, and is sent to the destination inter-node communicator 104. As a result, the waiting time until the data obtained by compressing the 8 kB portion 1001 is sent is the same as the time it takes to compress one 512 B portion 1002. In other words, the compression overhead time can be reduced by sequentially sending partial results during the compression process.

[0092] The dictionary range referenced when losslessly compressing 512B portion 1002 using the method shown in Fig. 9B is the range from the beginning of 8kB portion 1001 in which it is contained to itself (the range shown by 1004 in Fig. 10). This has the effect of improving the compression rate and compressing the amount of data to a smaller size (widening the apparent transmission bandwidth) because there are more character string candidates that can be used as a dictionary than if only 512B portion 1002 were used as the dictionary range.

[0093] Also, if the result of lossless compression of 512B portion 1002 individually is larger than 512B, a packet with the uncompressed 512B portion as its payload is sent instead of a packet with the compressed result as its payload. This has the effect of preventing the problem that compression increases the amount of data transferred and reduces the apparent transfer bandwidth. For example, 1005 and 1006 in Figure 10 show this. As shown in Figure 9B, literal characters are encoded into 9-bit patterns and the amount of data increases by 9 / 8, so 512B portion 1002, which has a very low frequency of copy symbols, is likely to exceed 512B when compressed.

[0094] FIG. 11 is a diagram showing the configuration of a PCI express packet transmitted and received by the inter-node communicator 104 when the inter-node transfer data is compressed and transferred.

[0095] PCIexpress packets generally consist of a TLP header and a payload. The TLP header contains a Length field that indicates the byte length of the payload, and an Address field that includes the destination address. The payload is the data itself to be sent. As explained in Figure 10, if the 512B portion can be reduced by lossless compression, the data is transferred in a compressed state, but if it cannot be reduced, the data is transferred in plain text.

[0096] A packet that sends 512B of plaintext data is called a plaintext packet, and a packet that sends less than 512B of compressed data is called a compressed packet. The payload of a plaintext packet is set to 512B of plaintext data 1103 (corresponding to 1005 and 1006 in Fig. 10), and the Length field 1101 of the TLP header is set to "512". The payload of a compressed packet is set to compressed data 1106 (corresponding to 1003 in Fig. 10) of 508B or less, and the Length field 1104 of the TLP header is set to the byte length of 1106. The reason why the size of the compressed data 1106 is set to 508 bytes or less is because the payload size of a PCI Express packet is specified as a multiple of 4 in principle. Therefore, if the compression result is 509 to 511 bytes, it is transferred as a plaintext packet.

[0097] In the plaintext packet and the compressed packet, the Address field 1102 or 1105 of the TLP header includes the following items: a destination memory address 1114, a device number 1111, a compression circuit ID 1112, and a compression state discrimination flag 1113.

[0098] The destination memory address 1114 is the destination address when the payload data is stored in the cache memory 106 in the destination storage node 101. In the case of a compressed packet, it is the address where the 512B data after decompression is stored. In other words, when an 8kB portion is sent in 16 packets, the 16 destination memory addresses 1114 included therein are assigned values ​​at 512B intervals.

[0099] The device number 1111 is a unique number for identifying the inter-node communication device 104 mounted in the storage node 101 to which the plaintext / compressed packet is transferred. The hub device 110 in FIG. 1 detects the device number 1111 included in the Address of the TLP header of the received packet and switches the transfer destination of the packet. The compression circuit ID 1112 is an identifier of the compression circuit. The compression circuit ID 1112 is a unique number for identifying which of the multiple compression circuits 212 in the inter-node communication device 104 that transmitted the packet performed the compression process on the payload of the packet.

[0100] The inter-node communicator 104 that received the packet performs decompression processing of the payload of the packet that includes the same compression circuit ID 1112 in the address of the TLP header, using the common decompression circuit 222. As shown in 1004 in Fig. 10, the dictionary range referred to in the compression processing of the 8kB part 1001 is shared by multiple 512B parts 1002, so it is necessary to use the same decompression circuit 222 to correctly decompress the data compressed by these.

[0101] The compression state discrimination flag 1113 is information for the decompression circuit 222 to identify whether the packet is a plaintext packet or a compressed packet. When this flag is "1", the packet is treated as a compressed packet and the payload is decompressed, but when this flag is "0", the packet is treated as a plaintext packet and the payload is bypassed from the decompression process.

[0102] In this embodiment, 512B of data is obtained from the payload of a plaintext packet, and 512B of data is obtained by decompressing the payload of a compressed packet by the decompression circuit 222. That is, data is always transferred in units of 512B to the reception output buffer 223 in Fig. 2. As described in the explanation of Fig. 1, in a PCI express connection in which the CPU 103 is the RootComplex and the inter-node communicator 104 is the EndPoint, the maximum payload size of a write transfer packet to the cache memory 106 is 512B.

[0103] By simply attaching a TLP header (where the address of the cache memory 106 is set in the Address field) to each data held in the reception output buffer 223 in 512 B units, a packet that can be directly written and transferred to the cache memory 106 can be formed, so the transmission process of PCI express is made more efficient. If data is held in units other than 512 B, a data size counter and a data division and concatenation circuit are required, but this embodiment can omit them.

[0104] 12 is a flow chart showing the compression process of the compression circuit 212 of the inter-node communicator 104 for compressing an 8 kB portion when the conditions are met for compressing and transferring inter-node transfer data. First, the index N is initialized to 0 (1201), and the process proceeds to step 1202.

[0105] In step 1202, the Nth 512B (the first is 0) is selected from the 16 512B that make up the 8 kB portion to be compressed. Then, for the selected 512B, dictionary compression is performed (1203) by including the history range (512B x N) in the dictionary. It is checked whether the size of the compressed result is 508B or less (1204). If this is true, the compressed result is transferred to the transmission output buffer 213 to be used as the payload (1205). If step 1204 is false, the original plaintext (512B) is transferred to the transmission output buffer 213 to be used as the payload (1206).

[0106] After step 1205 or 1206, N is incremented by 1 (1207). It is checked whether N is 16 (1208). If N is 16, the compression process is completed. If N is not 16, the process returns to step 1202, and compression process continues for the next 512B. At that time, the dictionary range used in dictionary compression in step 1203 includes the (N-1) 512B processed up to that point.

[0107] FIG. 13 is a flow chart showing the process in which the inter-node communicator 104 decompresses the 8 kB portion compressed by the compression circuit 212 using the decompression circuit 222 and transfers the decompressed portion to the cache memory 106.

[0108] First, the inter-node communicator 104 receives a packet (1301), and checks whether one of the decompression circuits 222 has been assigned to the compression circuit ID in the TLP header of the received packet (1302). If it has been assigned, the process proceeds to step 1304. If it has not been assigned, the process proceeds to step 1303, where it assigns one of the decompression circuits 222 and proceeds to step 1304.

[0109] In step 1304, the payload of the received packet is input to the allocated decompression circuit 222. Then, it is checked whether the compression state discrimination flag in the TLP header of the received packet is ON (compressed state) (1305). If it is OFF (plain text state), the process proceeds to step 1307. If it is ON, the decompression circuit 222 decompresses the payload (decrypts and expands the plain text by referencing a dictionary) to restore 512 B of plain text (1306), and the process proceeds to step 1307.

[0110] In step 1307, a packet for memory write is constructed from 512 B of plaintext (the decompressed result or the payload itself) and transferred to the cache memory 106. Then, it is checked (1308) whether a total of 8 kB of plaintext obtained from the received packet with this compression circuit ID has been transferred. If 8 kB has been transferred, the allocation of the decompression circuit 222 is released and the process is completed. If 8 kB has not been transferred, the process returns to receiving the subsequent packet (1301) and continues. At this time, the dictionary range referenced in the plaintext expansion in step 1306 includes all plaintext expanded since the allocation of the decompression circuit.

[0111] As described above, according to this embodiment, the communication data is compressed and expanded only when the communication band between storage nodes becomes a bottleneck in the data read or data write performance in the storage system, thereby making it possible to avoid wasting the compression and expansion processing time. Also, the compression and expansion is performed only on the compressible communication data (data other than RAID parity and already compressed data), making it possible to avoid wasting the compression and expansion processing time.

[0112] According to this embodiment, data is divided and compressed based on the maximum payload size of the write packet from the communication device to the cache memory, so that the write packet can be easily configured and data transfer can be made more efficient. Also, by sharing the dictionary range in dictionary compression of a plurality of divided data, the compression ratio can be improved (data volume can be further reduced), the apparent communication band between storage nodes can be further expanded, and data read / write performance can be improved.

[0113] (6) Other embodiments Another example of a control method in which the CPU 103 enables / disables compression by the inter-node communicator 104 will be described with reference to FIG. 14. This example is realized by changing step 804 in the flowchart of FIG. 8 to a conditional expression of "Is transmission output speed by the inter-node communicator - CPU performance < threshold value?" Here, the transmission output source is the inter-node communicator of the storage node storing the read data, and the transmission output destination is the inter-node communicator of the storage node that received the read command. Note that this condition may be further added without changing the above step 804. FIG. 14 shows the change in the transmission output speed by the inter-node communicator 104 and the read performance of the storage node 101 in this example. The CPU performance can be calculated from the CPU operating rate according to a specified formula.

[0114] When the frequency of issuing read commands from the host 108 increases, the CPU operating rate rises and so does the CPU performance; however, if compression of transfer data by the inter-node communicator 104 is disabled, the upper limit of the transmission output speed is 5.0 GB / s, so the difference with the CPU performance decreases. If it falls below the threshold, the process transitions to step 805, and the proportion of transfer data for which compression is enabled increases, thereby increasing the transmission output speed. If the difference with the CPU performance widens as a result, it exceeds the threshold and the process transitions to step 806, and the proportion of transfer data for which compression is enabled decreases. In other words, by changing to this conditional expression, the transmission output speed by the inter-node communicator 104 is maintained at a speed that is higher than the CPU performance by the threshold amount.

[0115] 14 shows the change in transmission output speed in this example. Until the transmission output speed reaches 10 GB / s, which is the speed when transfer data compression is always enabled, the bottleneck in read performance is CPU performance. In this control method, the inter-node communicator 104 does not enable compression of transfer data more than necessary, so that the occurrence of compression overhead time can be reduced more than in the control method of the previous example, and unnecessary increases in the response time until the read data is returned to the host 108 can be further suppressed.

[0116] The above explanation can also be applied to the processing of a write command. In step 804, the conditional expression "Is the transmission output speed by the internode communication device - CPU performance < threshold value?" is also set for the processing of a write command. Here, the transmission output source is the internode communication device of the storage node that received the write command, and the transmission output destination is the internode communication device of the other storage node. When increasing the ratio of transfer data for which compression is enabled in the control method of this example, the following priority command control may be performed. For example, a parameter indicating whether or not to allow an increase in response time is added to the read / write command from the host 108, and the internode communication data generated in the processing of a command for which the parameter is set to "No" is not selected as a target for enabling compression as much as possible. This makes it possible to minimize the deterioration of the response time for commands for which the host 108 does not want the response time to be deteriorated.

[0117] The present invention is not limited to the above-described embodiment, but includes various modified examples. For example, the above-described embodiment has been described in detail to clearly explain the present invention, and is not necessarily limited to those including all of the configurations described. In addition, it is possible to replace a part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment. In addition, it is possible to add, delete, or replace a part of the configuration of each embodiment with another configuration.

[0118] In addition, the above-mentioned configurations, functions, processing units, etc. may be realized in part or in whole by hardware, for example, by designing them as integrated circuits. In addition, the above-mentioned configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as the program, table, file, etc. that realizes each function can be stored in a memory, a recording device such as a hard disk or SSD (Solid State Drive), or a recording medium such as an IC card or SD card.

[0119] In addition, the control lines and information lines shown are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. In reality, it can be considered that almost all components are connected to each other. [Explanation of symbols]

[0120] 100...storage system, 101A, 101B...storage nodes, 103A, 103B...CPU, 104A, 104B...inter-node communication devices, 105A, 105B...data storage media, 106A, 106B...cache memories, 108A, 108B...hosts, 110...hub devices, 212...compression circuits, 222...decompression circuits.

Claims

1. A storage system including a plurality of storage nodes, Each storage node of the plurality of storage nodes comprises: A processor for processing external instructions; A drive for storing data; a communicator for transmitting data to or receiving data from another storage node; the communication device includes a compression circuit that losslessly compresses data before transmitting the data, and a decompression circuit that decompresses the compressed data after receiving the data, in response to a read command for reading data of a specified size to an external device, if a predetermined condition is satisfied, the communication device of the first storage node compresses data stored in the drive of the first storage node using the compression circuit, and transmits the compressed data to the communication device of the second storage node; the communication device of the second storage node decompresses the received data using the decompression circuit; The second storage node outputs the decompressed data to an outside.

2. 2. The storage system according to claim 1, The condition includes the size specified in the read command being equal to or greater than a threshold.

3. 2. The storage system according to claim 1, A storage system, wherein the condition includes that the data indicated by the read command is stored on the drive in an uncompressed state.

4. 2. The storage system according to claim 1, The condition includes an operating rate of the processor of the second storage node exceeding a threshold value.

5. 2. The storage system according to claim 1, A storage system, wherein the condition includes a difference between the data transmission output speed from the communication device of the first storage node to the communication device of the second storage node minus the processing performance of the read command by the processor of the second storage node being below a threshold value.

6. A storage system including a plurality of storage nodes, Each storage node of the plurality of storage nodes comprises: A processor for processing external instructions; A drive for storing data; A cache memory; a communicator for transmitting data to or receiving data from another storage node; The communication device includes a compression circuit that losslessly compresses data before transmission, and a decompression circuit that decompresses the compressed data after reception, in response to a write command for writing received data of a specified size from an external device, if a predetermined condition is satisfied, the communication device of the first storage node compresses data in the cache memory of the first storage node using the compression circuit and transmits the compressed data to the communication device of the second storage node; a communication device of the second storage node, the communication device decompresses the received data by the decompression circuit, and the second storage node stores the decompressed data in the cache memory of the second storage node.

7. 7. The storage system according to claim 6, The condition includes the size specified in the write command being equal to or larger than a threshold value.

8. 7. The storage system according to claim 6, The condition includes that the data in the cache memory is not RAID parity calculated from the received data.

9. 7. The storage system according to claim 6, The condition includes an operating rate of the processor of the first storage node exceeding a threshold value.

10. 7. The storage system according to claim 6, A storage system, wherein the condition includes a difference between the data transmission output speed from the communication device of the first storage node to the second storage node minus the processing performance of the write command by the processor of the first storage node being below a threshold value.

11. A storage system including a plurality of storage nodes, Each storage node of the plurality of storage nodes comprises: A cache memory; a communicator for transmitting data to or receiving data from another storage node; The communication device includes a compression circuit that losslessly compresses data before transmission, and a decompression circuit that decompresses the compressed data after reception, the communication device of the first storage node divides data into a plurality of partial data by using a maximum payload size of a write packet from the communication device to the cache memory as a division unit, compresses the partial data by the compression circuit, and transmits the resulting packet as a payload to the communication device of the second storage node; a communication device of the second storage node that receives the packet expands the payload of the received packet using the expansion circuit, constructs a write packet including the expanded payload, and transfers it to the cache memory of the second storage node.

12. 12. The storage system according to claim 11, A storage system, wherein the compression algorithm used by the compression circuit is dictionary compression, and the range of the dictionary referenced in the dictionary compression includes the plurality of partial data compressed by the compression circuit.

13. 12. The storage system according to claim 11, the communication device includes a plurality of compression circuits and a plurality of decompression circuits; the communication device of the first storage node adds an identifier of a compression circuit used to a packet to be transmitted to the communication device of the second storage node; The communication device of the second storage node selects one common decompression circuit among the plurality of decompression circuits for a plurality of packets including the identifier.

14. 12. The storage system according to claim 11, the communication device of the first storage node adds a flag indicating that the payload is in a compressed state to a packet to be transmitted to the communication device of the second storage node; A storage system, wherein the communication device of the second storage node determines whether to decompress the payload by the decompression circuit according to the flag.