[javagroups-users] Re: Big messages

[<Tak...@in...>]

Thank you for your response,

I tried frag_size program and I got 65507 bytes, that means that my frag_size 
has to beset to 64507 bytes. Actually, I run my tests (those that don't stop) 
with a frag_size=60000 bytes. Here is my protocol stack :
 "UDP(bind_addr=" + bindAddress + ";"+
            "ucast_send_buf_size=64000;"+
            "ucast_recv_buf_size=64000;"+
            "mcast_send_buf_size=64000;"+
            "mcast_recv_buf_size=64000):"+
            "PING:"+
            "MERGE2:"+
            "FD(timeout=20000):"+
            "VERIFY_SUSPECT(timeout=300000):"+
            "pbcast.NAKACK
(retransmit_timeout=600,1200,2400,4800;max_xmit_size=60000):"+
            "pbcast.STABLE(desired_avg_gossip=2000):"+
            "FRAG(frag_size=60000):"+
            "pbcast.GMS(join_timeout=3000;join_retry_timeout=2000)";

Note that I have increased VERIFY_SUSPECT timeout and FD timeout to minimize 
message exchange, I consider my environment close and secure (cluster 
environment)and failure probability is very low.

I changed frag_size and max_xmit_size to 8K,and my tests could finish. Do you 
thing that the problem is due to the increase of retransmission requests when 
fragmentation size is important?.
Takoua

Quoting Bela Ban <be...@ya...>:

> I just realized I didn't answer your email...
> 
> Can you post your config file ? You should run frag_size (in the CVS) to 
> determine your fragmentation size, and then set it (minus 1000 bytes) in 
> FRAG.frag_size and pbcast.NAKACK.max_xmit_size
> 
> Takoua ABDELLATIF wrote:
> 
> > Hi,
> > I tested JGroups with big messages (100K, 1 Megabytes). I noticed that 
> > even
> > with small configuration (4 nodes) and small burst of messages (10) , the
> > test lasts one hour and messages don't reach all the node members. How do
> > you explain this problem and are there any workaround.
> > Thank you in advance for your help.
> > Takoua
> >
> 
> -- 
> Bela Ban
> http://www.jgroups.org
> Cell: (408) 316-4459
> 
> 
>