Flow isolation method and device based on CXL Fabric

Abstract

The invention provides a traffic isolation method and device based on CXL Fabric, and the method comprises the steps: dividing an entrance cache and an exit cache into a request channel and a response channel in a management assembly of a switch; and analyzing the CXL management message routed to the management component, determining a destination address and a message type of the CXL management message, routing the CXL management message to the entry cache or the exit cache based on the destination address, and routing the CXL management message to a corresponding request channel or response channel based on the message type. According to the technical scheme of the invention, the interlocking of the request and the response when the management message is burst is avoided.

Description

Technical Field

[0001] The present application relates to the field of switching networks, and in particular to a traffic isolation method and device based on CXL Fabric. Background Art

[0002] In CXL (Compute Express Link, cache coherent interconnect protocol) switch chips, routing management is crucial, especially the management of routing rules in the multi-topology structure of the CXL3.1 protocol. CXL topology is divided into HBR (hierarchy based routing) network and PBR (port based routing) network. Figure 1 The dashed boxes indicate locations outside and inside, respectively. HBR networks use the traditional PCIe tree topology; PBR offers a more flexible, multi-topology approach, even including ring networks. HBR is based on the PCIe tree topology, with routing relying on hierarchical relationships. PBR supports flexible topologies (such as rings and meshes) and uses direct port addressing for routing, making it more suitable for complex network architectures. Summary of the Invention

[0003] The purpose of this application is to provide a traffic isolation method and device based on CXL Fabric, aiming to solve the interlocking problem that may exist when a CXL switch uses a single channel design to process sudden proxy management instructions.

[0004] According to a first aspect of the present application, a traffic isolation method based on CXL Fabric is provided, comprising:

[0005] In the management component of the switch, both the ingress cache and the egress cache are divided into request channels and response channels;

[0006] Parsing a CXL management message routed to the management component, determining a destination address and a message type of the CXL management message, routing the CXL management message to the ingress cache or the egress cache based on the destination address, and routing the CXL management message to a corresponding request channel or response channel based on the message type.

[0007] The CXL Fabric-based traffic isolation method of this application distinguishes management messages into request and response messages at the chip inlet, and separates the request and response channels in the management component. This avoids the interlocking of management requests and management responses within the chip management component in the event of a sudden increase in management message traffic without significantly increasing the burden on the switching component.

[0008] In an optional embodiment, routing the CXL management message to the ingress cache or the egress cache based on the destination address, and routing the CXL management message to a corresponding request channel or response channel based on the message type, further includes:

[0009] If the destination address of the current CXL management message is a management component and the message type is a request message, the current CXL management message is routed to the first request channel of the ingress cache.

[0010] If the destination address of the current CXL management message is a management component and the message type is a response message, the current CXL management message is routed to the first response channel of the ingress cache.

[0011] If the destination address of the current CXL management message is the topology administrator and the message type is a response message, the current CXL management message is routed to the second response channel of the egress cache.

[0012] If the destination address of the current CXL management message is the managed node and the message type is a request message, the current CXL management message is routed to the second request channel of the egress cache.

[0013] The method of the present application divides the ingress cache and egress cache into independent request and response channels, and implements corresponding routing management for different sources and destinations and different types of management messages. It can be widely used in management message burst scenarios where the routing management module performs in-band management through the switching interface.

[0014] According to a second aspect of the present application, a CXL Fabric-based traffic isolation device is provided, comprising:

[0015] A partitioning unit, configured to partition both the ingress cache and the egress cache into a request channel and a response channel in a management component of the switch;

[0016] a routing unit configured to parse a CXL management message routed to the management component, determine a destination address and a message type of the CXL management message, route the CXL management message to the ingress cache or the egress cache based on the destination address, and route the CXL management message to a corresponding request channel or response channel based on the message type.

[0017] In an optional embodiment, the routing unit is further configured to:

[0018] If the destination address of the current CXL management message is a management component and the message type is a request message, routing the current CXL management message to the first request channel of the ingress cache; or

[0019] If the destination address of the current CXL management message is a management component and the message type is a response message, routing the current CXL management message to the first response channel of the ingress cache; or

[0020] If the destination address of the current CXL management message is a topology manager (FM) and the message type is a response message, routing the current CXL management message to the second response channel of the egress cache; or

[0021] If the destination address of the current CXL management message is the managed node and the message type is a request message, the current CXL management message is routed to the second request channel of the egress cache.

[0022] Other features and advantages of the present application will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the present application. The purpose and other advantages of the present application can be achieved and obtained through the structures and processes indicated in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following is a brief introduction to the drawings required for use in the embodiments or related technical descriptions. It is obvious that the drawings described below are certain embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0024] Figure 1 FIG. 4 is a schematic diagram of a topological structure of a CXL according to related art.

[0025] Figure 2 2 is a schematic diagram of agent instruction routing in a CXL PBR network according to related art.

[0026] Figure 3 The diagram is a structural diagram of a switch with a built-in management component according to the related art.

[0027] Figure 4 It is a schematic diagram of the routing process of the proxy instruction according to the related art.

[0028] Figure 5 This is a schematic diagram of the interlocking state of the ingress cache and the egress cache according to the related art.

[0029] Figure 6 This is a flow chart of a traffic isolation method based on CXL Fabric according to the present application.

[0030] Figure 7 This is a schematic diagram of a CXL management message according to the present application.

[0031] Figure 8 This is a schematic diagram of the management components for request and response channel isolation according to this application. DETAILED DESCRIPTION

[0032] In order to make the purpose, technical solutions and advantages of the embodiments of this application more clear, the technical solutions in the embodiments of this application will be clearly and completely described below in conjunction with the drawings in the embodiments of this application. Obviously, the described embodiments are part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0033] There is a proxy instruction in the CXL PBR network, such as Figure 2 As shown, the proxy node needs to convert management messages and generate new request messages. After receiving responses from managed nodes, it generates new responses and returns them to the Fabric Manager (FM). Fabric management involves the unified coordination and control of the topology structure, routing rules, and device status within the CXL interconnection network, and involves the exchange of commands with switches and device nodes.

[0034] like Figure 3 As shown, each switch has a management component for managing routing information. Since the management interface can be any in-band interface, the management component is not directly exposed to the external interface, but communicates with each CXL interface through the access switch fabric.

[0035] The management process of the above agent instructions is as follows Figure 4 As shown:

[0036] First, the FM sends an original request to the switch's management component. The management component generates a proxy request and sends it to the managed node through the switch fabric. The managed node returns a response to the proxy request, which is forwarded back to the management component through the switch fabric. Upon receiving the proxy response, the management component generates a response to the original request and forwards it to the FM through the switch fabric.

[0037] The entry cache and exit cache of traditional management components have only one channel, so there will be Figure 5 The following situation is shown, that is, the FM request and the proxy response will enter the same ingress cache; and the proxy request and the FM response will be output from the same egress cache. Therefore, it may cause the ingress cache to be filled with the newly entered FM request message before the in-flight proxy response reaches the ingress cache, that is, Figure 5The problem of channel 1 back-pressing channel 3; on the other hand, the proxy response in flight is full of pipeline paths, which will back-press the managed nodes, that is, channel 3 will back-press channel 2; assuming that the head of the queue at the exit is channel 2, channel 2 will back-press channel 4, forming an interlocked state.

[0038] Based on the above analysis, the present application provides a traffic isolation method and device based on CXL Fabric. Based on the communication between the management component and the in-band interface outside the chip through the switching component, a special routing method is customized to combine address routing with type routing, and requests and responses are routed to different destinations of the management component. Furthermore, two cache channels are implemented in the management component to separate request instructions from response instructions, support burst management requests, and avoid request and response interlocking. At the same time, there is no need to divide physical channels on ordinary switching interfaces, thereby avoiding increasing the switching burden.

[0039] See also Figure 6 Flowchart, illustratively, the traffic isolation method based on CXL Fabric provided by this application includes:

[0040] Step 601: In the management component of the switch, both the ingress cache and the egress cache are divided into a request channel and a response channel.

[0041] The management component is connected to the switching network through a switching fabric, so that any external CXL interface can route management messages to the management component through a specific routing method.

[0042] The ingress cache and egress cache of the management component are divided into independent request channels and response channels. Specifically, the ingress cache can be divided into a first request channel and a first response channel. The first request channel is used to cache original request messages sent by the FM. The first response channel is used to cache response messages to proxy request messages returned by the managed node. The egress cache can be divided into a second request channel and a second response channel. The second request channel is used to cache proxy request messages to be sent to the managed node. The second response channel is used to cache response messages to the original request messages to be sent to the FM.

[0043] When the chip ingress performs routing judgment, if the message needs to be routed to the management component, the management message is parsed. Figure 7 As shown, the CXL management message is exemplarily an MCTP (Management Component Transport Protocol) message, and the TO field in the figure distinguishes whether it is a request message or a response message. For example, 0 represents a response message and 1 represents a request message.

[0044] Step 602: Parse the CXL management message routed to the management component, determine the destination address and message type of the CXL management message, route the CXL management message to the ingress buffer or egress buffer based on the destination address, and route the CXL management message to the corresponding request channel or response channel based on the message type.

[0045] When routing management messages at the chip entrance, the request instructions and response instructions are routed to the request channel and response channel of the management component respectively according to the aforementioned message parsing, thereby avoiding loop interlocking between the request message and the proxy response message.

[0046] Illustratively, if the destination address of the current CXL management message is a management component and the message type is a request message, the current CXL management message is routed to the first request channel of the ingress cache.

[0047] Illustratively, if the destination address of the current CXL management message is a management component and the message type is a response message, the current CXL management message is routed to the first response channel of the ingress cache.

[0048] Illustratively, if the destination address of the current CXL management message is FM and the message type is a response message, the current CXL management message is routed to the second response channel of the egress buffer.

[0049] Illustratively, if the destination address of the current CXL management message is a managed node and the message type is a request message, the current CXL management message is routed to the second request channel of the egress buffer.

[0050] By dividing the ingress cache and egress cache into independent request and response channels, corresponding routing management can be implemented for different sources and destinations, and different types of management messages. This can be widely used in management message burst scenarios where the routing management module performs in-band management through the switching interface. Figure 8 As shown, for example, channels 1 and 3 are separated, and channels 2 and 4 are separated. There is no need to implement two queues at the ingress of the chip's external interface port, because the FM and the managed node are not on the same ingress port.

[0051] It can be seen that the traffic isolation method based on CXL Fabric provided in this application has the following advantages compared with related technologies:

[0052] Management messages are divided into request and response messages at the chip entrance, and the request and response channels are separated in the management component. This avoids the interlocking of management requests and management responses within the chip management component in the scenario of management message bursts without significantly increasing the burden on the switching component. This can be widely used in management message burst scenarios where the routing management module performs in-band management through the switching interface.

[0053] Accordingly, in a second aspect, the present application provides a traffic isolation device based on CXL Fabric, comprising:

[0054] A partitioning unit, configured to partition both the ingress cache and the egress cache into a request channel and a response channel in a management component of the switch;

[0055] The routing unit is configured to parse the CXL management message routed to the management component, determine the destination address and message type of the CXL management message, route the CXL management message to the ingress cache or egress cache based on the destination address, and route the CXL management message to the corresponding request channel or response channel based on the message type.

[0056] In an optional embodiment, the routing unit is further configured to:

[0057] If the destination address of the current CXL management message is the management component and the message type is a request message, then route the current CXL management message to the first request channel of the ingress cache; or

[0058] If the destination address of the current CXL management message is a management component and the message type is a response message, then the current CXL management message is routed to the first response channel of the ingress cache; or

[0059] If the destination address of the current CXL management message is a topology manager (FM) and the message type is a response message, then the current CXL management message is routed to the second response channel of the egress cache; or

[0060] If the destination address of the current CXL management message is the managed node and the message type is a request message, the current CXL management message is routed to the second request channel of the egress cache.

[0061] The CXL Fabric-based traffic isolation device provided in this application has at least the following advantages over related technologies:

[0062] Management messages are divided into request and response messages at the chip entrance, and the request and response channels are separated in the management component. This avoids the interlocking of management requests and management responses within the chip management component in the scenario of management message bursts without significantly increasing the burden on the switching component. This can be widely used in management message burst scenarios where the routing management module performs in-band management through the switching interface.

[0063] It is understood that the structures, names, and parameters described in the above embodiments are merely examples. Those skilled in the art may also make readily conceivable combinations and adjustments to the structural features of the above embodiments as needed, and should not limit the concept of this application to the specific details of the above examples.

[0064] Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A traffic isolation method based on CXL Fabric, characterized in that: include: In the management component of the switch, both the ingress cache and the egress cache are divided into request channels and response channels; Parsing a CXL management message routed to the management component, determining a destination address and a message type of the CXL management message, routing the CXL management message to the ingress cache or the egress cache based on the destination address, and routing the CXL management message to a corresponding request channel or response channel based on the message type.

2. The traffic isolation method based on CXL Fabric according to claim 1, characterized in that: The step of routing the CXL management message to the ingress cache or the egress cache based on the destination address, and routing the CXL management message to a corresponding request channel or a response channel based on the message type, further includes: If the destination address of the current CXL management message is a management component and the message type is a request message, the current CXL management message is routed to the first request channel of the ingress cache.

3. The traffic isolation method based on CXL Fabric according to claim 1, characterized in that: The step of routing the CXL management message to the ingress cache or the egress cache based on the destination address, and routing the CXL management message to a corresponding request channel or a response channel based on the message type, further includes: If the destination address of the current CXL management message is a management component and the message type is a response message, the current CXL management message is routed to the first response channel of the ingress cache.

4. The traffic isolation method based on CXL Fabric according to claim 1, characterized in that: The step of routing the CXL management message to the ingress cache or the egress cache based on the destination address, and routing the CXL management message to a corresponding request channel or a response channel based on the message type, further includes: If the destination address of the current CXL management message is the topology administrator and the message type is a response message, the current CXL management message is routed to the second response channel of the egress cache.

5. The traffic isolation method based on CXL Fabric according to claim 1, characterized in that: The step of routing the CXL management message to the ingress cache or the egress cache based on the destination address, and routing the CXL management message to a corresponding request channel or a response channel based on the message type, further includes: If the destination address of the current CXL management message is the managed node and the message type is a request message, the current CXL management message is routed to the second request channel of the egress cache.

6. A traffic isolation device based on CXL Fabric, characterized in that: include: A partitioning unit, configured to partition both the ingress cache and the egress cache into a request channel and a response channel in a management component of the switch; a routing unit configured to parse a CXL management message routed to the management component, determine a destination address and a message type of the CXL management message, route the CXL management message to the ingress cache or the egress cache based on the destination address, and route the CXL management message to a corresponding request channel or response channel based on the message type.

7. The CXL Fabric-based traffic isolation device according to claim 6, characterized in that: The routing unit is further configured to: If the destination address of the current CXL management message is a management component and the message type is a request message, the current CXL management message is routed to the first request channel of the ingress cache.

8. The CXL Fabric-based traffic isolation device according to claim 6, characterized in that: The routing unit is further configured to: If the destination address of the current CXL management message is a management component and the message type is a response message, the current CXL management message is routed to the first response channel of the ingress cache.

9. The CXL Fabric-based traffic isolation device according to claim 6, characterized in that: The routing unit is further configured to: If the destination address of the current CXL management message is the topology administrator and the message type is a response message, the current CXL management message is routed to the second response channel of the egress cache.

10. The CXL Fabric-based traffic isolation device according to claim 6, characterized in that: The routing unit is further configured to: If the destination address of the current CXL management message is the managed node and the message type is a request message, the current CXL management message is routed to the second request channel of the egress cache.

Images