Hi there,
We have a client which uses Winflector to access our application (ERP), and we are experiencing some abrupt loss of application performance when the concurrent devices go higher than 18 (or so). They have licensed to allow up to 30 concurrent devices.
As we can tell, the application struggles to draw the client region image. When this issue appears, the application will draw its components in "chunks", and some intermediate "refresh flicks" appear, something that would not happen (or be perceptible) when the application is not run durinng this "heavy" load. Interestingly, other aspects of application's performance are apparently not affected (e.g.: a query to the database and row processing will take the same amount of time, looking at the logs).
Here are some information about this scenario:
- The server has 8 processors (Intel Xeon E5620 @ 2.4GHz) and 32Gb RAM;
- OS Version is Windows Server 2016 Standard Version 1607;
- Winflector Version is v3.9.6.7
- When this issue occurs, during the higher load of users, CPU is max 50-60% in use, and RAM memory will not overcome 50%; Measuring disk points to no bottleneck as well;
- Each user normally spawns multiple sessions (since each screen in the ERP application runs in its own process). When this scenario hits, session count may be as high as 100;
- The database (PostgreSQL) used by the ERP is installed in the same virtual machine as Winflector;
- They use Winflector authentication. That is, all sessions are concentrated in a single Windows/AD user profile;
- If we open the TS/RDP session of this user (which also runs the Winflector Server and concentrate all the connections) during the slow moments, we note that windows in general inside this RDP session also takes way longer to be drawn (e.g. opening a task manager will take up to 1 min to be fully drawn, with its tabs and list of processes);
- If we open another RDP session using another user, performance looks ok (task manager will show up in an expected time-frame: 1-2s);
Have you experienced anything like that? Any advices? Any kind of log we should look into?
I suspect that the concentration of all these users in a single session may be the culprit - something like exhausting the number of handles/gdi objects/threads/whatever of the user.
I am trying to convince my client to create a Windows/AD user per device and change the authentication mode. But since this will give some work and effort is not guaranteed to be effective, I am asking you if this is the right direction or should I look anything else?
