[Spread-users] Message reception time

[Yair Amir]

Hi Alberto,

A few months ago we released a 2-pager report that contain some
throughput and latency measurements for Spread. If you read it
it will give you some idea about what kinds of latencies to expect
depending on the kind of service you use. Agreed or lower should
get you latency of around 2-3 ms depending on the size of the message
(this was measured in a 20 nodes network.

"The Spread Toolkit: Architecture and Performance", CNDS-2004-1
http://www.cnds.jhu.edu/publications/

Or directly:
http://www.cnds.jhu.edu/pub/papers/cnds-2004-1.pdf

Cheers,

       :) Yair.

On Sunday, November 14, 2004 10:07 PM
Ryan Caudy rcaudy at gmail.com wrote:

Ryan> It is possible that the Java implementation is slower.  If all else
Ryan> fails, I would try the same thing with a simple C client and server,
Ryan> and see if the measurements are different.

Ryan> Your Spread_Segment definition seems fine, assuming you're just using
Ryan> two machines, with clients and servers connecting locally to a daemon
Ryan> on their machine.  With only two machines, there shouldn't be any
Ryan> significant token latencies to worry about slowing down the delivery,
Ryan> but Yair or Jonathan may have more ideas about this.

Ryan> Do you know if IP multicast is working properly on your network?  You
Ryan> could use spmonitor to see what the _retrans counts it reports are at
Ryan> each daemon... I wouldn't expect to see many retransmissions at all. 
Ryan> If IP multicast isn't working, you might try declaring the segment
Ryan> with your network's broadcast address instead.

Ryan> I assume ping latencies on your network are much much lower than 90
Ryan> ms.  On our usual test LAN networks, they tend to be less than 1 ms
Ryan> (0.1 ms - 0.2 ms).

Ryan> When you report 90 ms, is that the average of multiple measurements? 
Ryan> If so, how many, and are you restarting you configuration for each, or
Ryan> not?  If you are averaging a relatively small number, I could imagine
Ryan> that the strange performance is caused by a timeout in Spread's token
Ryan> rotation protocol, which seeks to minimize traffic due to the token on
Ryan> inactive Spread networks.

Ryan> Let me know if any of these ideas help.

Ryan> Cheers,
Ryan> Ryan

Ryan> On Sat, 13 Nov 2004 19:13:09 +0100 (CET), Alberto Martinez
Ryan> <albertinho_2000 at yahoo.es> wrote:
>> Hi!
>>     That's just the times I need. However my project
>> is designed using Java. I've been looking at openais
>> and is implemented in C language. Could it be that
>> Java implementation is slower than C while sending
>> messages???
>> 
>> 
>> 
>> Thanks!
>> 
>> ______________________________________________
>> Renovamos el Correo Yahoo!: ¡100 MB GRATIS!
>> Nuevos servicios, más seguridad
>> http://correo.yahoo.es
>> 
>> _______________________________________________
>> Spread-users mailing list
>> Spread-users at lists.spread.org
>> http://lists.spread.org/mailman/listinfo/spread-users
>>