I am trying to find out if there is anyone else using multiple client managers on a load balancer and if we could have a discussion regarding such a setup. We have this in our envrionment and it does work, however sometimes I see oddities with in the client - even if I log directly into one of my client managers, by-passing the load balancer completely.
Looking forward to hearing from anyone with a similar setup.
Hi there, we are still in the middle of our upgrade but we have similar set up.
We have two CMs and use Cisco GSS to load balance and offload SSL handling to so that our CM's don't have to be bogged down with SSL.
Let me know what oddities you see specifically. We are running on Linux 6, using Java 1.7.
I myself have encountered many caching issues on the admin side (workgroups and user management) and have applied 3 different patches since we created our upgrade environment to no avail. I have had to update rows in the database behind the scenes since cache do not match database (in 5.3 we NEVER update the database directly). I am just hoping that these quirks will be fixed before we go live.
I am also conflicted about the DSP setting on Transporter and the Apple app since it bypasses the load balancer and forces you to indicate a specific CM. Admittedly i am still playing around at this early stage.
Our environment is also still test/evaluation and I have applied may patches as well, more today.
Did you do any stress testing using your load balancer and response from CM? We noticed a slow down of functions such as inserting in to schedule and filtering, although our environment is quite different than most; we did our test with 32 people signed in. What is the highest number of connections you will have at one time in your environment?
Have you noticed in the Master status/connections multiple logins for the same user? We have seen this and it can be very disconcerting - I have queried the Cache DB when I see this and these records are there (and they are NOT from users exiting incorrectly) the records will hang around in the database for a time but will ultimately disappear we asked if we would experience a performance hit with these 'rogue' connections we were told we would not. Although response is slow regardless.
I'm not too concerned about Transporter being connected to a specific CM as we are not giving the right to use transporter to end users.
We haven't done a formal load testing yet. But we have conducted training with 10 - 15 users and would experience performance sometimes people just getting kicked out. At the time though we were still finagling with server configuration settings for threads, cache and java heap. During training too what seemed to impact performance more was the browser and version, and workstation memory. For a web app, TES 6.1 uses a LOT of memory so there were some suggested tweaks that had to be done on some machines. We have more consistent performance on IE 9 (than IE 10), and Firefox 18+ was faster.
I have installed Jconsole and am planning to conduct load testing sometime this month where i could monitor server resource from Jconsole to get an idea where bog down really is.
There is a TES 6.1 presentation doc that sort of outlines approx sizing based on concurrent users - we fit in the 10 -20 concurrent users given that our CM has 12GB RAM (allocated 10GB to TES) and 4 core. So hopefully with two CMs we can have upto 40. We are open to adding more mem if needed. So what tools do you use for Load testing? Also I wonder if it was possible to automate Load Testing like with Jmeter or something. For now we just plan on getting 20 users together to perform set functions at the same time and measure that with 1 CM up and me looking in Jconsole.
I also notice the multiple connections, even see myself twice even when I only have 1 browser session. Not having had a chance to dig to deeply i was chalking it up to maybe that I didn't exit properly last time so that it has to wait 30 minuts to totally clear me (with time to live setting) out but haven't really tested formally. I will see about adding that to my testing. I would also think it would impact system in that some folks may cause the number of concurrent connections allowed to be exceeded if you have too many of these lying around but since they are not doing anything perhaps not performance as much? It will be good to confirm though that a single web session really only takes 1 connection and won't potentially spawn another somehow.
Our implementation being highly distributed with workgroups having full autonomy over their jobs - means every team needs to have access to Transporter. We have over 15 teams. Which is scary since that also takes resources from the CM - and also from a management and training perspective. We also haven't load tested the Transporter server interms of number of concurrent users that can be running Transporter on it at one time. Nor have any clear idea yet how we'll manage the mapping files. But we are also opening up Transporter to be installed by teams on their team server (only that they have to patch themselves) - I really wonder if we should just restrict Transporter access to just the servers our team manage and no one else can install it anywhere else.