Attendees:
Arnie
Nate
Michael
Tim
Jimmy
Chris UE Unicon
Mark Unicon
Michael Pollack Unicon
Khalil
Michael Morris
Agenda:
The agenda for tomorrow will be catching up on what everyone is doing. We need to give TIm a complete overview of what's been going on, and catch each other up on our varied action items.
Discussion:
Khalil: Cloud services. Had the impression that AWS was set up. Working on Azure. Ask Nate to contact Mike ? to get things going. What is the role of TSG in all of this? From the Azure side we need TSG to have an account "owner". A paperwork function, they won't have a direct role, but someone has to own it. Nate is fine with that approach. Nate will email Larry and cc list about getting resources provisioned. The grant may not be allocated yet, because we think Larry might be waiting to hear from I2. We need to experiment with both to see if one or the other is better. We want to keep both in our pockets in case we run into an issue with one or the other. The bottom line goal: Cloud resources next week.
Jimmy: Using jmeter to exercise APIs because that appears to be the best way to test. Ever since the switch over from SOAP to API's the performance has improved. Improved logging and hibernate caching. Screen shot emailed is the result of this morning's testing. He's working on logging and messaging changes. Move over to PostGRES. Various testing to be done. Very straightforward to write these test. Working on this next week. Need to exercise GUI, working with UI folks. Found a toolkit that allows testing a UI using junit. Main focus for next week or more.
Resin discussion. Might have slowdown issues. First test to determine how the application engine manages it's own memory and push that to the limit. Resin's strength will be on high load with multiple systems. We may need to make a trade-off decision between feature rich vs. speed. Unicon has been maintain artifacts to inform configuration manager.
Chris: Needs a direction to go. Are PSU resources available to collaborate with Chris?
Action Items:
Nate follow up with Larry and cc list and particularly Tim to get cloud services rolled out.
Unicon resin tests across a cluster. Prepare to port from Unicon environment to our cloud VM environment.
Unicon:
Test Resin in a single server environment and test Jetty behind a proxy in a single server environment. All artifacts perserved to assist in setting up VMs. LDAP server on a separate server. Nate will email the entire action item list to the list.
Nate and Chris will coordinate on UE. Jimmy will work with Renee to see if we can get PSU resources. Minimal modifications to CPR desirable. Make CPR skinned to CommIT needs.
Testing approach below. In all instances I would like to track metrics on throughput, latency(max, mean, min), error rate, CPU load, and memory utilization. As much concurrency as possible and as much reality in terms of directory/database size as possible.
Initial work that begins now:
1) Deploy, in a local environment controlled by Unicon,
1a) a VM containing an LDAP directory with dummy user data(running 389)
1b) a VM containing Resin running the IdP
1c) a VM containing ha proxy and Jetty running the IdP
The IdP will be configured for LDAP authentication and resolution/release of a few attributes. Each configuration will be tested using the Grinder framework maintained by the Shibboleth project.
2) Chris will begin work on a stand-in logo and general theme for the initial phase of the project using CommIT as the public name. Full artistic license since he knows more about this stuff than we do.
2a) Jimmy will ask Renee to see if some PSU GUI resources would be available for a minimal workload of getting the GUI codebase/database
3) Nate will begin the finalization of procurement of VM resources, owned by Internet2's tech support group(TSG), by sending out a pair of emails. Project management will chase TSG and cloud service providers to ensure things get done.
4) Jimmy will migrate the PSU CPR to use Postgres as a backing store and tune interfaces appropriately and test it on a server with 4 CPU's
4a) Benchmarking on GUI performance should begin, to be initially done by the PSU GUI team; Jimmy saw a framework that might be usable
Next steps:
5) Deploy, in each cloud environment(AWS and Azure),
5a) a VM containing an LDAP directory with dummy user data(presumably 389, and presumably only 1 if it's not an immediate bottleneck)
5b) VM's containing 2 or N(N guesstimated based on preliminary numbers) clustered instances of Resin running the IdP
5c) a VM containing ha proxy(presumably only 1 if it's not an immediate bottleneck) with layer 7 stickiness and Jetty running the IdP on 2 or N(N guesstimated based on preliminary numbers) VM's.
Same test harness run.
Other next steps, especially GUI work and CPR work, TBD -- it's too late here tonight.