[ome-users] Improving webclient performance

[Aleksandra Tarkowska (Staff)]

Hi Paul

On 11 May 2016, at 16:31, Paul van Schayck <paul at vanschayck.nl<mailto:paul at vanschayck.nl>> wrote:

Thanks for the help. I probably forgot to mention the crucial bit that
I'm running OMERO 5.1.4. Could that be the biggest influence, looking
at all the changes you've made to OMERO.web organization between 5.1
and 5.2?

YES! There were many changes to web, in particular data loading.
Also, bug making unnecessary calls to the server on each request, was fixed in 5.2.2 https://github.com/openmicroscopy/openmicroscopy/pull/4474.
Sorry about that.

I've done some more testing as you proposed using `ab` and the P-TRE
test image. Also setting public access to not use bSession.

$ ab -n 10 -c 1
"http://omero.local/webclient/render_thumbnail/size/96/201/?server=1&bsession=ae766ef0-7520-4878-822c-419eb3faf6b8"
Connection Times (ms)
             min  mean[+/-sd] median   max
Connect:        1    1   0.1      1       1
Processing:   268  319  18.6    323     338
Waiting:      268  319  18.5    323     338
Total:        269  320  18.5    324     338

$ ab -n 10 -c 1
"http://nanoscopy.unimaas.nl/webclient/render_thumbnail/size/96/201/"
Connection Times (ms)
             min  mean[+/-sd] median   max
Connect:        1    1   0.1      1       1
Processing:   391  412  11.2    415     420
Waiting:      390  412  11.2    415     420
Total:        391  412  11.2    416     421

As you can see, actually bSession is faster than (by about 100ms) than
the public user. The render_thumbnail is about 100ms faster than the
webclient/ index call I tested with yesterday. But overall, the 320 ms
average is pretty far from the 100 to 150 you report.

I've attached the requested outputs. Note that the amount of errors in
Blitz.log is most likely due to a lot of experimenting I've done in
the past.

Did you try to increase number of gunicorn workers by changing  --workers WORKERS?

Did you also try to tune nginx?
It may not make the biggest difference, but is still possible that tweaking worker_processes, worker_connections, worker_rlimit_nofile, using poll, etc. will speed it up.

Did you try to deploy 5.2.3 on a similar VM to see if overal performance is better?
Alternatively, to free up resources you could deploy OMERO.web with nginx on a different VM then OMERO server. If you need any help with that let us know.


Thanks

Ola
Software Engineer
Open Microscopy Environment
University of Dundee

The University of Dundee is a registered Scottish Charity, No: SC015096
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openmicroscopy.org.uk/pipermail/ome-users/attachments/20160511/1bf833f0/attachment.html>