Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

RDMA buffer dynamic allocation method based on iterative approximation

An RDMA and dynamic allocation technology, applied in the computer field, can solve the problems of RDMA buffer memory occupation, insufficient RDMA buffer, memory waste, etc., and achieve the effect of reducing the number of RDMA buffers, reducing memory consumption, and high bandwidth

Active Publication Date: 2021-03-02
JIANGNAN INST OF COMPUTING TECH
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0013] The disadvantage of this method is: the RDMA initiator needs to wait for the passive party to allocate the RDMA buffer before performing RDMA operations, which introduces additional message delays
[0016] The disadvantage of this method is that a large number of RDMA buffers need to be allocated in advance to prevent insufficient RDMA buffers; and in the case of low load, this will cause the RDMA buffer to occupy a large amount of memory, resulting in memory waste

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • RDMA buffer dynamic allocation method based on iterative approximation
  • RDMA buffer dynamic allocation method based on iterative approximation
  • RDMA buffer dynamic allocation method based on iterative approximation

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0033]Embodiment: A method for dynamic allocation of RDMA buffers based on iterative approximation, including the following steps:

[0034]S1. During initialization, the RDMA initiator makes the following settings: record the number of buffers requested to be allocated as rbuf_thresh, the initial value is set to 4, the time of rbuf_thresh expansion is recorded as epoch_start, the initial value is set to the current time, rbuf_thresh before rbuf_thresh expansion Denoted as Rmin , The initial value is set to 2, and the expanded rbuf_thresh after rbuf_thresh is recorded as Rmax , The initial value is set to 4, R after rbuf_thresh is expandedmax Reduce to Rmin The time is denoted as K, initialized as K=, Where C is a constant coefficient;

[0035]S2. The RDMA initiator checks the list of RDMA buffers stored locally. This RDMA buffer list contains all available RDMA buffers allocated by the RDMA passive party to the RDMA initiator, and judges whether the RDMA passive party has available RDMA ...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses an RDMA buffer dynamic allocation method based on iterative approximation, which comprises the following steps: S1, during initialization, an RDMA initiator performing the following settings: the number of buffers requested to be allocated is recorded as rbuf_thresh, the initial value being set as 4, the time during rbuf_thresh expansion is recorded as epoch_start, the initial value is set as the current time, and the rbuf_thresh value before rbuf_thresh expansion is recorded as Rmin, the initial value being set as 2, rbuf_thresh, after expansion, being marked as Rmax,with the initial value as 4, wherein the time for reducing Rmax to Rmin after rbuf_thresh expansion is recorded as K, K is initialized to be equal to K (please refer to specification for the formula),and C is a constant coefficient; S2, the RDMA initiator checking a locally stored RDMA buffer area list, and judging whether the RDMA passive party has an available RDMA buffer area or not; and S3, after the RDMA passive party receives an RDMA buffer area allocation request sent by the RDMA active party, allocating a new RDMA buffer area according to the rbufthresh in the request, and notifying the RDMA active party of the new RDMA buffer area. According to the method and the device, dynamic allocation of the RDMA buffer areas is realized, the number of the pre-allocated RDMA buffer areas canbe remarkably reduced while the high bandwidth of the network is ensured, and the memory consumption is reduced.

Description

Technical field[0001]The invention relates to an RDMA buffer dynamic allocation method based on iterative approximation, and belongs to the field of computer technology.Background technique[0002]The exchange-based interconnection network architecture, including IB, provides data transmission types such as Send / Receive, RDMA Write, and RDMARead. The Send / Receive operation is similar to the send / receive semantics in socket programming, and the data specified by the sender can be written into the receive buffer specified by the receiver. RDMA Write, RDMA Read and other operations (hereinafter collectively referred to as RDMA operations) can write the data specified by the sender into the remote receiving buffer specified by the sender without the participation of the remote host, thereby providing direct access to the upper software The programming semantics of remote memory has become the most common operation type of programming interfaces such as IBVerbs.[0003]The usage flow of RDMA...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06F15/173
CPCG06F15/17331
Inventor 何王全漆锋滨陈淑平李祎王申卢德平周慧霖钱炜
Owner JIANGNAN INST OF COMPUTING TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products