activemq-users mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Charles Anthony <charles.anth...@hpdsoftware.com>
Subject RE: ThroughPut : Small vs Large Messages
Date Thu, 22 Jun 2006 13:15:50 GMT
Oh well - I tried that but to no avail.

I've just dumped the size of our messages (ObjectMessage, so I just
serialized the object to a byte[] and took the length - I'm sure it's a bit
out, as I imagine there is some overhead for the wire-format - but not by a
huge amount) and 98% are > 2K anyway, so I've just set the default messages
size in the diagnostic tool to 2k.

Cheers,

Charles.

-----Original Message-----
From: James Strachan [mailto:james.strachan@gmail.com]
Sent: 22 June 2006 14:19
To: activemq-users@geronimo.apache.org
Subject: Re: ThroughPut : Small vs Large Messages


Yeah

On 6/22/06, Charles Anthony <charles.anthony@hpdsoftware.com> wrote:
> Thanks - would that be the TCP_NODELAY jobby ?
>
> -----Original Message-----
> From: James Strachan [mailto:james.strachan@gmail.com]
> Sent: 22 June 2006 11:30
> To: activemq-users@geronimo.apache.org
> Subject: Re: ThroughPut : Small vs Large Messages
>
>
> It could be the Nagler kicking in - i.e. it might take a little while
> for the TCP buffers to be flushed to the network for smaller messages.
> You can tune TCP to enable/disable waiting for complete tcp buffers
> before actually sending it to the network - its a throughput v latency
> tradeoff setting.
>
> On 6/22/06, Charles Anthony <charles.anthony@hpdsoftware.com> wrote:
> > Hi,
> >
> > We use AMQ (4) as the client/server transport in our system. For some of
> our
> > clients, we host the application at our site, and though connect
remotely
> > via VPN. Sometimes, we (well, usually they) have trouble setting up the
> VPN
> > and setting firewall configs etc - so I am just knocking together a
little
> > noddy diagnostics program - i.e. connect to the AMQ server, create a
temp
> > queue, create a producer, create a consumer, send and receive a load of
> > messages.
> >
> > All well and good.
> >
> > I thought I'd go a bit crazy, then, and try and work out message
> throughput
> > based on message size (i.e. send & reveive A x byte[B] => bytes sent =
> A*B,
> > bytes-per-second= A*B/num-secs) - not as a real measurement, but more as
> > wavy-hand type uide.
> > I've turned async send off for this diagnostic thingy, and messages are
> sent
> > NON_PERSISTENT.
> >
> > Here's the weird thing.
> >
> > If send small messages ( < 1300 ish bytes), I am getting roundtrip of
> approx
> > 200 ms
> > If send larger messages ( > 1300 ish bytes - say 2048), I am getting
> > roundtip of < 15 ms
> >
> >
> > Is there a reasonable explanation for this ? Maybe compression kicking
in
> ?
> > I'm just a bit befuddled.
> >
> > Cheers,
> >
> > Charles.
> >
> >
> > ___________________________________________________________
> > HPD Software Ltd. - Helping Business Finance Business
> > Email terms and conditions: www.hpdsoftware.com/disclaimer
> >
> >
> >
>
>
> --
>
> James
> -------
> http://radio.weblogs.com/0112098/
>
>
> ___________________________________________________________
> HPD Software Ltd. - Helping Business Finance Business
> Email terms and conditions: www.hpdsoftware.com/disclaimer
>
>
>


-- 

James
-------
http://radio.weblogs.com/0112098/


___________________________________________________________
HPD Software Ltd. - Helping Business Finance Business
Email terms and conditions: www.hpdsoftware.com/disclaimer 



Mime
View raw message