Re: Re : serveriron predictor

Hi Rubens,

You are dead on  correct on DSR issue with response time. Ah well, I can pretty much say the only options I have are;

1. use weighted predictor
2. use server response time weights
3. or used either of the two above in combination of "max-tcp-conn-rate".

Thanks,
Reynald



-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Rubens Kuhl
Sent: Wednesday, June 24, 2009 12:50 PM
To: Load Balancing Mailing List
Subject: Re: [load balancing] Re : serveriron predictor

Reynald,

Haven't tried that one, because DSR (Direct Server Return) makes the
response times available only to the health checks, not to every
connection.

As you have a dynamic parameter I think it's still a good to thing to
use max-tcp-conn-rate per server in order to prevent too much "boat
swinging".

Also, Foundry docs list a "smooth factor" for use with response-time
that is likely to boost stability.


Rubens



On Tue, Jun 23, 2009 at 10:35 PM, Reynald
Mahinay<[email protected]> wrote:
> Hello Surya and Rubens,
>
> You both have a good point. In relation, I look-up the net and found serveriron can do something like response-time + weighted servers. It's like combining the two predictor you guys have suggested.
>
> Have you tried that setup before?
>
> Regards,
> Reynald
>
> -----Original Message-----
> From: [email protected] [mailto:[email protected]] On Behalf Of Rubens Kuhl
> Sent: Tuesday, June 23, 2009 10:57 PM
> To: Load Balancing Mailing List
> Subject: Re: [load balancing] Re : serveriron predictor
>
>>   least-conn
>>   least-local-conn
>>   least-local-sess
>>   local-weighted
>>   response-time
>>   round-robin
>>   weighted
>
> I suggest the weighted predictors. Heavy loads are prone to cause
> instability with response-time or least-conn, with the server that was
> most responsive been killed by the next wave of connection setups that
> will all go to that server.
>
> Unless the balancer can override this behavior with a max new
> connection per second rate, I advise to use static balancing, weighted
> by the effective http power(measured with ab or your favorite http
> load tool), not simple CPU clock math.
>
>
> Rubens
> _______________________________________________
> lb-l mailing list
> [email protected]
> http://vegan.net/mailman/listinfo/lb-l
> Searchable Archive: http://vegan.net/lb/archive
> http://lbdigest.com Load Balancing Digest
> http://lbwiki.com Load Balancing Wiki
>
> _______________________________________________
> lb-l mailing list
> [email protected]
> http://vegan.net/mailman/listinfo/lb-l
> Searchable Archive: http://vegan.net/lb/archive
> http://lbdigest.com Load Balancing Digest
> http://lbwiki.com Load Balancing Wiki
>
_______________________________________________
lb-l mailing list
[email protected]
http://vegan.net/mailman/listinfo/lb-l
Searchable Archive: http://vegan.net/lb/archive
http://lbdigest.com Load Balancing Digest
http://lbwiki.com Load Balancing Wiki

_______________________________________________
lb-l mailing list
[email protected]
http://vegan.net/mailman/listinfo/lb-l
Searchable Archive: http://vegan.net/lb/archive
http://lbdigest.com Load Balancing Digest
http://lbwiki.com Load Balancing Wiki