All sub-jobs in this test exited successfully.

This page contains the results of the hlt2 bandwidth test with the production streaming configuration. Scroll down to see:

• Summary of main results,
• Details of the streaming configuration,
• Links to other html pages produced by this test,
• Bar charts of rate and bandwidth for each WG within each stream (HLT2 only),
• A pie chart of all lines split by WGs (HLT2 and sprucing only),
• Information about the input sample,
• Stacked histograms of all lines, split by WG, of rate/bandwidth metrics.

The total bandwidth (rate) was measured to be 11.46 GB/s (255.69 kHz).

The streaming configuration (i.e. which lines went to each stream) can be found in JSON format here.
This streaming configuration is our current set of lines to be used in the next data-taking period.
"DstData" is the raw bank to which reconstructed information (candidates, other reconstructed tracks etc.) are saved.
The "DstData bandwidth" is therefore the bandwidth counting only that raw bank.
The total event size (and total bandwidth) count all raw banks (incl. DstData, and detector raw banks if present) in the file.
NB: In real data-taking, raw banks are now left uncompressed when writing, and then the whole file is compressed afterwards. We account for this compression by multiplying event sizes and bandwidths by a scaling factor to give accurate per-stream bandwidths. The scaling factor is calculated for each file in the test as: '(size of the compressed file) / (size of the uncompressed file)'.

Further results can be found in the links below:

• A single rate/bandwidth table featuring every trigger line in all streams
• Rate/bandwidth tables for each stream, with 1 row per trigger line
• Rate/bandwidth tables for each stream, split also by WG, with 1 row per trigger line
• Jaccard similarity and overlap matrices between streams
• Bar charts as below for DstData bandwidth
• Similarity and overlap matrices between WGs within each stream
• List of lines in "Other" category
• PersistReco and ExtraOutput info for all lines in all streams

See: RTA & DPA Workflow for reference figures regarding bandwidth.

Rate within each stream:

"TotalInclusive" is the physical rate/bandwidth of the stream. "SumWGs" is the simple arithmetic sum of all bars except "TotalInclusive".
The difference between the two bars gives us information about the degree of WG-by-WG overlap.

Note: The WG bars in the HLT2 Turbo stream correspond almost exactly to the output streams of the Sprucing passthrough of Turbo.
Furthermore, this means the "SumWGs" rate bar of HLT2 turbo is approximately equal to the total physical rate of Turbo post-sprucing.

Bandwidth within each stream:

"TotalInclusive" is the physical rate/bandwidth of the stream. "SumWGs" is the simple arithmetic sum of all bars except "TotalInclusive".
The difference between the two bars gives us information about the degree of WG-by-WG overlap.

Note: The WG bars in the HLT2 Turbo stream correspond almost exactly to the output streams of the Sprucing passthrough of Turbo.
Furthermore, this means the "SumWGs" bandwidth bar of HLT2 turbo is approximately equal to the total physical bandwidth of Turbo post-sprucing.

Rates for a WG within a stream are calculated by counting the number of events saved to that stream in which at least 1 of that WG's lines fired.
Bandwidths for a WG are calculated by summing the event size of all events saved to the stream in which at least 1 of that WG's lines fired.

The number of selection lines per working group:

"Other" category contains those lines with a parsed name that doesn't belong to any known WG.
To make lines properly categorized, one should follow the naming convention - name of lines should start with `Hlt2/Spruce[WG]_`.

Input sample information:

• Config file: /home/lblocal/jenkins/perf/workspace/nightly-builds/tests/build/Moore/Hlt/Hlt2Conf/tests/options/bandwidth/hlt2_bandwidth_input_2024.yaml
• Input rate: 1000 kHz
• Number of interactions per bunch crossing (ν): 7.6
• Radius of VELO opening: 5.1 mm

Stacked histograms of all lines, split by WG, of rate/bandwidth metrics:
The total distributions are shown as a stacked histogram, split into several histograms of WGs.
The distributions per WG is attached in the html page linked above.
Total event size is calculated from summing all raw banks in the file (including DstData).
Where appropriate, the DstData raw bank size and DstData bandwidth are calculated from summing only the DstData raw bank.