[Openais] FW: Query about openais performance

Hegde, Ramesh (STSD) rameshh at hp.com
Thu Feb 8 21:09:52 PST 2007


Hi Steve,

Thank you very much. Now I understand it better and it is not a
bottleneck anymore. I hope the special configuration is  a runtime
configuraton(openais.conf).  Can you please provide me the configuration
changes for 200 node cluster?

Regards
Ramesh 

-----Original Message-----
From: Steven Dake [mailto:sdake at redhat.com] 
Sent: Friday, February 09, 2007 12:23 AM
To: Hegde, Ramesh (STSD)
Cc: openais at lists.osdl.org
Subject: Re: FW: Query about openais performance

Ramesh,

The code is compiled to support 384 processors (nodes).  It would
require some special timeout configurations to actually get to 384 nodes
and performance would be sub par.  I have simulated 128 nodes at maximum
before without any problems.

The big issue is that latency of the protocol is directly related to the
size of the ring.  Specifically latency is 1/2 token rotation time.  A
token takes about 300usec per node if that node has no messages to
transmit so latency can be high. 300usec*200 nodes is 60 msec just for a
token rotation with no messages to transmit on the ring.  During this
time new queued messages are delayed from transmit until that node holds
the token.  This averages out to about half the token rotation time, so
latency = 30 msec on an empty ring.  Hope that explains it.

You will most certainly have to change the default timing parameters
which I can help with.  If you have a 200 node cluster to test on I can
help with the configuration.

Regards
-steve

On Thu, 2007-02-08 at 10:49 +0530, Hegde, Ramesh (STSD) wrote:
> Hello,
>  
> Can you please help me in understanding the below performance data.
>  
> Regards
> Ramesh
> 
> 
> ______________________________________________________________________
> From: Hegde, Ramesh (STSD)
> Sent: Friday, February 02, 2007 4:05 PM
> To: openais at lists.osdl.org
> Cc: Hegde, Ramesh (STSD)
> Subject: Query about openais performance
> 
> 
> Hello,
>  
> In OpenAis EVS performance I found that there is a limitation on 
> openais. i.e
>  
> " The current openais executive supports 16 processors, however , 
> support for more processors is possible by changing defines in the 
> openais executive"
>  
> Can some one explain me this better? Does it mean that only 16 nodes 
> are supported in a cluster environment? or  16 processor is  for given

> node or machine.
>  
> I am mainly concerned about the OpenAis Event Service in particular in

> a cluster environment of around 200 nodes.. Do you anticipate any 
> issues or bottleneck against my performance requirements?
>  
> Your inputs will help me very much .
>  
> Regards
> Ramesh
>  
> Ramesh Hegde
> 
> Software Engineer, Hewlett Packard
> System Technology and Software Division 3, Edward Road, Bangalore - 
> 560 052. India.
> +91 80 2205 3838 - Direct
> +91 80 2235 3604 - Fax
> Jabber Id : rameshh at jabber.hp.com
> +hp = everything is possible
>  





More information about the Openais mailing list