Baunsgaard commented on PR #1869:
URL: https://github.com/apache/systemds/pull/1869#issuecomment-1651717766
Initial in memory performance test of the Compression and processing speed,
The results show that we can update and apply a compression scheme at high
rate throughput.
The numbers indicate Input throughput and output throughput,
As an example Sum is performed at 7.9GB/s input and produce output at 8.33
KB/s
```
WriteTest Repetitions: 100 ConstMatrix 1000, 1000, 1.0)
Warmup Sum task Single Thread, [ 0.443, 0.462, 0.479,
0.567, 26.832], 7.94 GB/s In, 8.33 KB/s Out
Warmup Sum task Parallel , [ 0.422, 0.441, 0.511,
0.561, 3.409], 14.52 GB/s In, 15.22 KB/s Out
Compression In Memory Single Thread, [ 55.041, 56.349, 56.940,
58.169, 232.053], 128.03 MB/s In, 22.66 MB/s Out
Compression In Memory Parallel , [ 15.457, 33.681, 47.115,
49.519, 54.410], 189.31 MB/s In, 33.51 MB/s Out
Update & Apply Scheme, [ 18.761, 20.947, 21.263,
21.495, 39.948], 358.98 MB/s In, 63.54 MB/s Out
Apply Scheme, [ 8.682, 8.751, 8.815,
8.928, 11.203], 857.06 MB/s In, 151.71 MB/s Out
Update & Apply Scheme Parallel, [ 3.099, 3.567, 3.652,
3.759, 8.134], 2.02 GB/s In, 366.32 MB/s Out
Apply Scheme Parallel, [ 1.204, 1.513, 1.561,
1.600, 4.642], 4.72 GB/s In, 856.16 MB/s Out
```
FYI @mboehm7 , Do you think this is the right direction?
Following tests will confirm if this also works when writing to disk.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
