[From nobody Wed Jul 13 02:52:55 2011
Message-ID: <453634B6.4060102@cs.jhu.edu>
Date: Wed, 18 Oct 2006 10:05:42 -0400
From: Yair Amir <yairamir@cs.jhu.edu>
User-Agent: Thunderbird 1.5.0.7 (Windows/20060909)
MIME-Version: 1.0
To: pierre.garcia@inrialpes.fr
Subject: Re: [Spread-users] Performance evalution, bursty application traffic
	: are my results normal ?
References: <453622F1.1060107@aol.com>
In-Reply-To: <453622F1.1060107@aol.com>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Hi Pierre,

Your results are expected with your network configuration -
you get linear degradation if each machine is its own segment.

However, if all of them in the same segment, I believe you should
see no degradation if they all send and some much smaller degradation
to the one you have if one or few send. Otherwise, there may be
a problem with your setup which you may be able to track using the
spmonitor, and the spsend and sprecv programs provided with Spread.

I think this short paper shows what can be achieved on a fairly
similar configuration:
http://www.cnds.jhu.edu/pub/papers/cnds-2004-1.pdf

Cheers,

	:) Yair.

Pierre GARCIA wrote:
> Hi Spread Group,
> 
> I'm a new Spread User. I plan to use spread to safely multicast messages 
> (from 1Ko to 100Ko) on 20
> machines.
> 
> Description of the environment
> 
> Spread Version : 3.17.3
> Switch                : Fast Ethernet
> Nodes                 : - 2 x 900MHz Itanim2 cpu
>                              - 3 Gbytes memory
>                              - 2 x 36 Gbytes Hard Disks
>                              - OS : Red Hat Enterprise Linux AS release 
> 3 (Taroon Update 5)
> 
> 
> I have a bursty pattern to generate traffic. Thus, I tuned WATER_MARK 
> and MAX_SESSION_MESSAGES to increase performances.
> 
> I have one daemon/client per node and my spread.conf looks like this :
> 
> spread.conf
> --------------------------------
> Spread_Segment xx.88.97.255:4803 {
>    machine1 xx.88.97.61
> }
> Spread_Segment xx.88.97.255:4804 {
>    machine2 xx.88.97.62
> }
> Spread_Segment xx.88.97.255:4805 {
>    machine3 xx.88.97.63
> }
> Spread_Segment xx.88.97.255:4806 {
>    machine4 xx.88.97.64
> }
> Spread_Segment xx.88.97.255:4807 {
>    machine5 xx.88.97.65
> }
> Spread_Segment xx.88.97.255:4808 {
>    machine6 xx.88.97.66
> }
> Spread_Segment xx.88.97.255:4809 {
>    machine7 xx.88.97.67
> }
> Spread_Segment xx.88.97.255:4810 {
>    machine8 xx.88.97.68
> }
> Spread_Segment xx.88.97.255:4811 {
>    machine9 xx.88.97.69
> }
> Spread_Segment xx.88.97.255:4812 {
>    machine10 xx.88.97.70
> }
> Spread_Segment xx.88.97.255:4813 {
>    machine11 xx.88.97.71
> }
> Spread_Segment xx.88.97.255:4814 {
>    machine12 xx.88.97.72
> }
> Spread_Segment xx.88.97.255:4815 {
>    machine13 xx.88.97.73
> }
> Spread_Segment xx.88.97.255:4816 {
>    machine14 xx.88.97.74
> }
> Spread_Segment xx.88.97.255:4817 {
>    machine15 xx.88.97.75
> }
> Spread_Segment xx.88.97.255:4818 {
>    machine16 xx.88.97.76
> }
> Spread_Segment xx.88.97.255:4819 {
>    machine17 xx.88.97.77
> }
> Spread_Segment xx.88.97.255:4820 {
>    machine18 xx.88.97.78
> }
> Spread_Segment xx.88.97.255:4821 {
>    machine19 xx.88.97.79
> }
> Spread_Segment xx.88.97.255:4822 {
>    machine20 xx.88.97.80
> }
> 
> --------------------------
> 
> I have tried several other configurations (one segment per example), but 
> performances were worst.
> 
> My experience is minimal : each client sends ~20Mo of data. This amount 
> of data is divided in several messages (from 20000 messages of  1Ko to 
> 200 messages of  100Ko).
> 
> My RESULTS (Message size 10 Ko):
> 
> Nodes     Average Throughput (Mb/s)
> 2             83.93
> 3             45.02
> 4             30.33
> 5             22.85
> 6             18.3
> 7             15.29
> 10           10.22
> 15           6.57
> 20           4.85
> 
> For other sizes of message, the decrease is the same.
> 
> My questions are :
> 
>  1) Is it normal ? Did I make something wrong ?
> 
>  2) Is there other parameters in spread that can help me to increase 
> performances ?
> 
> If anyone can answer, you're welcome,
> 
> Pierre
> 
> _______________________________________________
> Spread-users mailing list
> Spread-users@lists.spread.org
> http://lists.spread.org/mailman/listinfo/spread-users

]