New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Reduction of RAW' data size via adjustment of approx. cluster member data types #39199
Conversation
+code-checks Logs: https://cmssdt.cern.ch/SDT/code-checks/cms-sw-PR-39199/31809
|
A new Pull Request was created by @abaty for master. It involves the following packages:
@jpata, @cmsbuild, @mandrenguyen, @clacaputo can you please review it and eventually sign? Thanks. cms-bot commands are listed here |
test parameters:
|
@cmsbuild please test |
+1 Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-315a48/27098/summary.html Comparison Summary@slava77 comparisons for the following workflows were not done due to missing matrix map:
Summary:
|
@perrotta @qliphy I want to draw your attention to this PR. This is not the "technical bug fix" that I understand was discussed in the ORP and is planned to be backported from 12_6_X. This PR effectively changes the compression level of the approximated clusters, and does have an effect on the relevant workflows (161 and 140.58) It could be considered a bug-fix in the sense that the default compression level was doing nearly nothing. In any case, this is needed for ion running, so we'll have to find a way to get it into 12_5_X. |
Technically it can be still be considered a bug fix, thus I don't see problems in backporting it to 12_5_X before 12_5_0. |
+reconstruction |
This pull request is fully signed and it will be integrated in one of the next master IBs (tests are also fine). This pull request will now be reviewed by the release team before it's merged. @perrotta, @dpiparo, @qliphy, @rappoccio (and backports should be raised in the release meeting by the corresponding L2) |
+1
|
@abaty @mandrenguyen @icali IIUC, the current strategy has a massive inefficiency for V0 reconstruction |
@slava77 Currently we are working on another raised issue regarding the backwards compatibility of this strategy, so we have not had time to critically examine the feasibility/accuracy of storing cluster shape info at HLT. We will look into this issue once the backwards compatibility is resolved. For the upcoming HI run, we have at the moment disabled the cluster shape filter so the massive V0 inefficiency is not expected. I agree that making a separate issue to track this for the future would be a good idea.
|
This PR is aimed at significantly reducing the size of the RAW' data format which uses the Approximated Cluster class, as the current implementation is not fully optimized. This is the subsequent PR for improving the format which is mentioned at the end of the first paragraph in the description of PR #38423.
The space savings is accomplished by changing the full-precision float class members of SiStripApproximatedCluster to integer data types having limited precision. In the case of the cluster width and average charge, values are rounded to the nearest whole number. For the barycenter, rounding to one decimal place was found to give a good tradeoff of tracking performance and space savings . To avoid using a floating point number, the barycenter is multiplied by 10 and stored as a 16-bit integer, and is converted back when regenerating the cluster collection from approximated clusters. This reduces the approximated cluster collection size by 37% over the current implementation, and the SiStrip detector payload by 45% compared to the conventional RAW format. A small loss (~0.5%) of inclusive tracks was observed, but largely disappears when applying analysis-type track selections. (More info on our tests at [1], which was discussed in our data format working group)
This PR also incorporates two suggestions from @makortel for cleaning up the approximated cluster code [2,3].
[1] https://twiki.cern.ch/twiki/pub/Main/HIDetectorReadout2020/ApproxClusters_integer10th_Aug25.pdf
[2] 8871eff
[3] 0492f1f
PR validation:
This PR can be tested using WF 140.58 and checking step2 to verify the smaller approx. cluster size. Step 3 can be used to check physics performance and tracking differences.
@icali @mandrenguyen