This topic always pops up when the time of the exams starts. And with every new thinlinc release i hope its addressed in one way or another.
i’m desperately hoping of a real loadbalancing.
We have about 70+ students at the exams starting their session rather simultaneously to a cluster of 10 thinlinc servers. Since there is no some sort of round-robin the students are distributed extremely uneven to the servers until the the penaltypoints or rating kicks in. this leads sometimes to very bad situations.
I really don’t know how deal withthis problem and this is a real problem for us!
Hello!
I have not tried this myself, but if you could try to modify the “existing_users_weight” setting.
In my (default) installation it is set to 4, try to set it to 1.
From that fine manual
/vsmserver/existing_users_weight
This parameter decides the importance of the amount of logged in users on a VSM agent host when calculating load balance parameters. A host with low load, but a lot of users, is generally more likely to get a higher load within short time when the users get active. For this reason, the load balance calculating code takes the number of users at a certain host into its calculation. The /vsmserver/existing_users_weight controls how important this factor is. A higher value of this parameter means the load balancing code will care less about a high number of users on a certain machine.
Note
This parameter should normally not be changed, unless when fine-tuning the load balancing.
2 Likes
@jean thanks for your feedback.
Load balancing is a tricky one, since there are a lot of potential scenarios to cater for. Ideally it would be a lot more configurable, and this is something we are discussing.
In the mean time, as @pewo mentioned above, there are some parameters which may help with your situation, and changing these would be worth a try. Another idea might be to group your agents into subclusters, and restrict each subcluster to a certain group. You could then assign your users evenly across these groups.
Let us know how you get on.
With the start of the new semester in post-covid-mode I could observe the outcome of the changes made to the thinlinc server (existing_user_weight = 2 , load_update_cycle = 5 ).
The users are now better distributed over the 10 thinlic-servers but not yet perfect!
thanks for your help
2 Likes
@jean thanks for the update. I’m glad to hear it’s an improvement!