CXL-based memory disaggregation technology opens up a new direction for big data solution frameworks​

0 0
Read Time:3 Minute, 45 Second


Determine 1. a comparability of the structure between CAMEL’s CXL resolution and traditional RDMA-based reminiscence disaggregation. Credit score: KAIST

A workforce from the Laptop Structure and Reminiscence Programs Laboratory (CAMEL) at KAIST offered a brand new compute categorical hyperlink (CXL) resolution whose instantly accessible, and high-performance reminiscence disaggregation opens new instructions for large information reminiscence processing. Professor Myoungsoo Jung stated the workforce’s know-how considerably improves efficiency in comparison with current distant direct reminiscence entry (RDMA)-based reminiscence disaggregation.

CXL is a peripheral element interconnect-express (PCIe)-based new dynamic multi-protocol made for effectively using and accelerators. Many enterprise information facilities and distributors are listening to it because the next-generation multi-protocol for the period of massive information.
Rising large information functions similar to , graph analytics, and in-memory databases require giant reminiscence capacities. Nonetheless, scaling out the through a previous reminiscence interface like double information charge (DDR) is proscribed by the variety of the central processing models (CPUs) and reminiscence controllers. Due to this fact, reminiscence disaggregation, which permits connecting a number to a different host’s reminiscence or reminiscence nodes, has appeared.
RDMA is a method {that a} host can instantly entry one other host’s reminiscence through InfiniBand, the generally used community protocol in . These days, most current reminiscence disaggregation applied sciences make use of RDMA to get a big reminiscence capability. In consequence, a number can share one other host’s reminiscence by transferring the information between native and distant reminiscence.

CXL-based memory disaggregation technology opens up a new direction for big data solution frameworks​

Determine 2. A efficiency comparability between CAMEL’s CXL resolution and prior RDMA-based disaggregation. Credit score: KAIST

Though RDMA-based reminiscence disaggregation gives a big reminiscence capability to a number, two crucial issues exist. First, scaling out the reminiscence nonetheless wants an additional CPU to be added. Since passive reminiscence similar to dynamic random-access reminiscence (DRAM), can’t function by itself, it must be managed by the CPU. Second, redundant information copies and software program cloth interventions for RDMA-based reminiscence disaggregation trigger longer entry latency. For instance, distant reminiscence entry latency in RDMA-based reminiscence disaggregation is a number of orders of magnitude longer than native reminiscence entry.
To deal with these points, Professor Jung’s workforce developed the CXL-based reminiscence disaggregation framework, together with CXL-enabled custom-made CPUs, CXL units, CXL switches, and CXL-aware working system modules. The workforce’s CXL system is a pure passive and instantly accessible reminiscence node that accommodates a number of DRAM twin inline reminiscence modules (DIMMs) and a CXL reminiscence controller. Because the CXL reminiscence controller helps the reminiscence within the CXL , a number can make the most of the reminiscence node with out processor or software program intervention. The workforce’s CXL swap permits scaling out a number’s reminiscence capability by hierarchically connecting a number of CXL units to the CXL swap permitting greater than tons of of units. Atop the switches and units, the workforce’s CXL-enabled working system removes redundant information copy and protocol conversion exhibited by standard RDMA, which might considerably lower entry latency to the reminiscence nodes.

In a check evaluating loading 64B (cacheline) information from reminiscence pooling units, CXL-based reminiscence disaggregation confirmed 8.2 instances greater information load efficiency than RDMA-based reminiscence disaggregation and even related efficiency to native DRAM reminiscence. Within the workforce’s evaluations for a giant information benchmark similar to a machine learning-based check, CXL-based reminiscence disaggregation know-how additionally confirmed a most of three.7 instances greater efficiency than prior RDMA-based reminiscence disaggregation applied sciences.
“Escaping from the standard RDMA-based reminiscence disaggregation, our CXL-based reminiscence disaggregation framework can present excessive scalability and efficiency for various datacenters and cloud service infrastructures,” stated Professor Jung. He went on to emphasize, “Our CXL-based reminiscence disaggregation analysis will convey a few new paradigm for reminiscence options that can lead the period of massive information.”

‘Memory disaggregation’ for large-scale computing made practical

Supplied by
KAIST

Quotation:
CXL-based reminiscence disaggregation know-how opens up a brand new route for large information resolution frameworks​ (2022, March 16)
retrieved 21 March 2022
from https://techxplore.com/information/2022-03-cxl-based-memory-disaggregation-technology-big.html

This doc is topic to copyright. Aside from any honest dealing for the aim of personal examine or analysis, no
half could also be reproduced with out the written permission. The content material is supplied for info functions solely.



Source link

Happy
Happy
0 %
Sad
Sad
0 %
Excited
Excited
0 %
Sleepy
Sleepy
0 %
Angry
Angry
0 %
Surprise
Surprise
0 %

Average Rating

5 Star
0%
4 Star
0%
3 Star
0%
2 Star
0%
1 Star
0%