Why might my batch analysis performance be slower in Columbus 2.9 when compared with the same batch analysis on the same data in Columbus 2.8.x?
There is a known issue in Columbus 2.9 which occurs under very specific circumstances and relates directly to the Processor configuration of the server hardware whereby a library inside the Columbus architecture incorrectly reports 0 NUMA nodes and 16 cpu sockets/packages. This means that Columbus will bind most of the analysis threads to a single cpu package, which in this case only contains a single cpu core, so that the remaining 15 cpu cores remained mostly unused. This is a result of changes made in the Columbus 2.9 release which are not present in prior versions of Columbus hence why this might impact Batch Analysis performance.
This issue has already been fixed and will be available in the next release of Columbus (Columbus 2.9.1). No confirmed release date at time of writing this FAQ.
If you believe that you are seeing this type of behavior we recommend that you contact the Imaging/Screening Technical support team using the following email address:
They will request additional information regarding the server configuration.
Prior to contacting the Technical Support team and in order to facilitate root cause analysis we recommend that you collate the following information.
For an overview of the server and processor configuration the output of the following commands will be required:
$ free -m
$ less /proc/cpuinfo
A copy of the Columbus log files after running the batch analysis. The logs can be retrieved in one of two ways.
via command line:
$ zip -r /tmp/columbus_logs.zip /var/log/columbus
via Columbus web UI (enter the following into the address bar of your chosen browser):
columbusip/help/error/all_logs.zip (replace 'columbusip' with the IP address for the server in question)
Additional logging is also required. The additional logging involves making a slight change to one of the files on the Columbus server as detailed below.
Change logging in the /var/lib/acapella/5.1.x/flags file:
The existing content of the flags file should read something like:
-log all 5 -log databundle 5 -log acc 6 -logfile 5.1.0/log/acc.log
Edit the flags file to read:
-log all 5 -log databundle 6 -log AAL 6 -log acc 6 -logfile 5.1.0/log/acc.log
i.e. increase the logging level of the ''databundle'' to 6 and add logging for the AAL code and set it to level 6.
Now re-run the batch analysis and collect the log files.