server full GC stops all clients

[yogi]

Hi Group,

I have an interesting issue related to GC in client server topology,

When one server embarks on a full GC, all clients and servers come to a
halt until the GC is complete. So, if a system had N servers and each
server hit a full GC every M seconds, there would be a global flatline
every M/N seconds.

For example, GC every 80 seconds and 20 servers averages 1 global
flatline every 4 seconds.

It is unclear why all clients and all servers stop on one server's full
GC.

Thanks in Advance.

Yogi

 

[Oliver Wong]

Hi Group,

I have an interesting issue related to GC in client server topology,

When one server embarks on a full GC, all clients and servers come to a
halt until the GC is complete. So, if a system had N servers and each
server hit a full GC every M seconds, there would be a global flatline
every M/N seconds.

For example, GC every 80 seconds and 20 servers averages 1 global
flatline every 4 seconds.

It is unclear why all clients and all servers stop on one server's full
GC.

Thanks in Advance.

How are the servers connected to each other, if at all? Do the servers
ever wait for data from a client?

- Oliver

 

[yogi]

Thanks for your reply Oliver.

Yes server does wait for data from client.
Servers are connected meanig .. they are running in same VM.

-Yogi

 

[Oliver Wong]

[post re-ordered]

Yes server does wait for data from client.
Servers are connected meanig .. they are running in same VM.

Perhaps running them in different VMs would solve the GC flatline
problem.

- Oliver

 

[Chris Smith]

Perhaps running them in different VMs would solve the GC flatline
problem.

Also, -Xincgc may help, if they need to be in the same VM.