The running time of the 'canneal' benchmark is quite unstable. #47
Replies: 1 comment 2 replies
|
Hi, Use of Pin in Sniper inherently brings in some randomness in the final result. It is also affected by OS activity to some extent, which adds to the entropy. Why the entropy is higher for Cannel than other benchmarks, I do not know. Cannel nevertheless is a highly memory-intensive benchmark. Sniper does model main memory capacity. By default it is 1024 MB I think, and you can change it in base config. Sniper doesn't have concept of virtual memory, so I don't know what will happen if this main memory gets exhausted. Since we run mostly sim-small and sim-medium benchmarks in HotSniper .. 1 GB is more than enough. -- Anuj |
2 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment


Uh oh!
There was an error while loading. Please reload this page.
Dear Professor,

When I run the canneal benchmark repeatedly under the same configuration, no matter what mapping policy or dvfs policy is used, the result of final runtime is unstable, there will be a relatively large difference in the final runtime. However, when I ran other benchmarks, under the same configuration, the running time was relatively stable, and there was no such situation as canneal.
I want to know why this happens? I searched for relevant information and learned that canneal is strongly related to main memory, and the Sniper in HotSniper does not seem to model the main memory capacity. Is this related to it?
All reactions