Data Caching

Results for Running on a 64 Core ARM machine

The machine had 4 NUMA nodes and each time we ran the server and client in different NUMA nodes. The server was configured to run at 2.4 GHz.

Results for Running on a 24 Core Intel machine (PARSA Cluster)

The machine had 2 NUMA nodes and each time we ran the server and client in different NUMA nodes. The server was configured to run at 2.5 GHz with the Intel turbo boost disabled.

Please note that the errors are normalized regarding the square root of RPS.

Compared Results from the Two Architectures

Done by Ali Asgari - memcached 1.5.9

Results from running on the x86 machine from PARSA cluster: Results from running on the x86 machine from IC cluster (Same with what used for Web Search):

Done by Siddharth Gupta - memcached 1.5.10

Results from running on Huawei ARM server and IC cluster - 4 cores 2GB: You can download the data for aarch64 or for x86.

Late Results for Qemu

I plotted for data caching on Qemu using this.

This is not a good graph because I ran it in 2 sessions for the left hand and the right hand but as the CPU utilization is continuous I think it's fine.

I used single threaded Qemu which was using only one host core for the server. I couldn't use the icount option for mrun as I mentioned here.