Skip to content
GitLab
Explore
Sign in
Register
Primary navigation
Search or go to…
Project
A
alltoall_benchmark
Manage
Activity
Members
Labels
Plan
Issues
Issue boards
Milestones
Wiki
Code
Merge requests
Repository
Branches
Commits
Tags
Repository graph
Compare revisions
Snippets
Build
Pipelines
Jobs
Pipeline schedules
Artifacts
Deploy
Releases
Package registry
Model registry
Operate
Environments
Terraform modules
Monitor
Incidents
Analyze
Value stream analytics
Contributor analytics
CI/CD analytics
Repository analytics
Model experiments
Help
Help
Support
GitLab documentation
Compare GitLab plans
Community forum
Contribute to GitLab
Provide feedback
Keyboard shortcuts
?
Snippets
Groups
Projects
Admin message
为了安全,强烈建议开启2FA双因子认证:User Settings -> Account -> Enable two-factor authentication!!!
Show more breadcrumbs
Ionizing
alltoall_benchmark
Commits
f5b96845
Commit
f5b96845
authored
1 year ago
by
Ionizing
Browse files
Options
Downloads
Patches
Plain Diff
[README.md] add test result from tuyy
parent
97951161
No related branches found
No related tags found
No related merge requests found
Changes
1
Hide whitespace changes
Inline
Side-by-side
Showing
1 changed file
README.md
+38
-0
38 additions, 0 deletions
README.md
with
38 additions
and
0 deletions
README.md
+
38
−
0
View file @
f5b96845
...
@@ -94,3 +94,41 @@ would be related to number of MPI ranks. Thus the more processes you used in `mp
...
@@ -94,3 +94,41 @@ would be related to number of MPI ranks. Thus the more processes you used in `mp
| node3 | 16 | 0.5 GB | 0.352 |
| node3 | 16 | 0.5 GB | 0.352 |
| node[1,2] | 8
\*
2 | 0.5 GB | 0.430 |
| node[1,2] | 8
\*
2 | 0.5 GB | 0.430 |
| node[3,4] | 8
\*
2 | 0.5 GB | 1.935 |
| node[3,4] | 8
\*
2 | 0.5 GB | 1.935 |
## Tuyy's Test of NAMD-EPC
Source on
[
GitHub
](
https://github.com/vtzf/MPI_Comm_Test
)
### 根据 Alltoall/Alltoallv 16 核测试结果:
1.
系统更新后,大规模通信(>=1GB)效率明显提升:
-
h3c 节点内通信效率提升 4 倍,sugon 节点内通信效率提升 3 倍
-
h3c 节点间通信效率提升 1.3 倍,sugon 节点间通信效率提升 4 倍
-
h3c 节点间通信效率在更新前后均优于节点内通信,不符合常理,仍需测试
2.
中等规模通信(64-512MB)效率上:
-
h3c 节点内通信效率略有下降,sugon 节点内通信效率变化不大
-
h3c 节点间通信效率变化不大,sugon 节点间通信效率提升 10-15 倍
3.
小规模通信(<=64MB)效率上:
-
h3c 节点内通信效率变化不大,sugon 节点内通信效率变化不大
-
h3c 节点间通信效率提升 2 倍,sugon 节点间通信效率提升 30-50 倍
### 根据 NAMD-EPC 32 核测试结果:
系统更新后,
Transpose 模块与一次
`MPI_Alltoallv`
大规模通信和多层 for 循环计算相关,
TimeProp 模块与 mkl 计算库和中等次数
`MPI_Iallgatherv`
的小规模通信相关,
SurfHop 模块与中等次数
`MPI_Allgather`
和大量次数
`MPI_Isend`
/
`Irecv`
的小规模通信相关,
1.
Transpose 模块上,
-
h3c 单节点计算效率基本不变,跨节点计算效率提升 1.5-2 倍,
-
sugon 跨节点计算效率大大提升 20-30 倍
2.
TimeProp 模块上,
-
h3c 单节点和跨节点计算效率基本不变
-
sugon 跨节点计算效率提升数倍
3.
SurfHop 模块上,
-
h3c 单节点和跨节点计算效率基本不变
-
sugon 跨节点计算效率大大提升 50-100 倍
This diff is collapsed.
Click to expand it.
Preview
0%
Loading
Try again
or
attach a new file
.
Cancel
You are about to add
0
people
to the discussion. Proceed with caution.
Finish editing this message first!
Save comment
Cancel
Please
register
or
sign in
to comment