Powered by Indico
v3.2.7
Present - Catherien Zwahlen (CZ) Lukasz Dutka (LD) and Matthew Viljoen (MV)
- Both dCache and Cyfronet EGI instances of HBP data/imaging software the
are still up and running since the Bari demonstration. LD reported
problems due to logs filling up a partition. To prevent this happening in
the future, CZ will see whether there is a way to reduce log verbosity in
the imaging service (installed as a Docker image), while LD will see
whether an automatic cleanup can be implemented.
- Catherine sent an email on 8 Dec with a list of http requests to the
imaging service simulating likely workflow that a user is likely to
follow when using the visualisation applet client.
- MV asked whether we can deploy the visualisation client on EGI
infrastructure for testing purposes. All present agreed that this is
would be good. The client is a JavaScript applet currently deployed
behind HBP firewall. It presents a GUI to end users to facilitate slicing
of a dataset. The applet then requests 'tiles' or small portions of an
image from the imaging service, then combines them to create the whole
image. To be deployed on EGI infrastructure, the visualisation applet
needs to be configurable to be able to connect to different sites.
-> HBP team will work to make the visualisation applet deployable to EGI
and configurable to use arbitrary FQDNs hosting the imaging service
- Performance requirements of the visualisation applet were discussed -
and how we can ensure our deployed instances of the imaging service and
dataset were serving requests sufficiently fast. CZ said that we should
aim for 8 tiling requests to the imaging service to be served within 1
second. We can also try multiple concurrent groups of 8 tiling requests
to determine the maximum achievable limit. Examples should use the
workflows in Catherine's email (mentioned above)
-> Write a script to simulate workflows from one (and more than one) user
to gather performance metrics by determining how quickly requests are
served. (LD)
- LD asked about obtaining larger datasets than the one we have. LD/MV
said that if open data access is a concern, we can protect access to the
imaging service via security, e.g. username and password - or via
integration with OpenID (currently used by HBP), during the testing phase.
CZ said this was possible, but there were some issues with OpenID within
the HBP that need to be resolved first. CZ also said she would look at an
alternative data provider to make a larger datasets deployable on EGI.
-> EGI to look at ways to protect the imaging service behind a username
and password to prevent open access
-> HBP team should identify an image provider internal to HBP and look at
ways for EGI to host the data on a non-open basis during testing (CZ)
DONM: Tue 22 Dec 10:30 (MV and LD - CZ sends her apologies)