Ultra low latency in cluster application

Trustin Lee tlee at redhat.com
Thu Feb 26 07:03:39 EST 2009


Hi David,

On Thu, Feb 26, 2009 at 1:56 AM, David Alves <davidralves at gmail.com> wrote:
>  The only way I can prevent the problem is by using plain old blocking
> Sockets, there the latency is minimal, but the throughput also plummets 10
> fold (dropping from 40.000 to 5.000 on my mac). What I find curious is that
> Oio in netty is supposed to be built over the old sockets right? If i use
> Oio the latency still increases...

This is really interesting.  Oio in Netty is a very thin layer built
on top of the old sockets, so it shouldn't involve any additional
latency.  If you used Oio on all nodes, then you should see the same
result with the plain sockets.  Otherwise, it means whether 1) there's
a bug in Netty or 2) there's a bug in your app.

Anyway, I'd appreciate your help if you could send me some
reproduceable code.  It might be working just fine in my machines, but
who knows? :)

— Trustin Lee, http://gleamynode.net/




More information about the netty-users mailing list