[SunRay-Users] Failover group load balancing on servers with lots of cores

William Yang wyang at tjhsst.edu
Tue Mar 16 16:02:57 EET 2010


That's an interesting point, I didn't think that CPU throttling was relevant
here, but under Linux at least, that seems to be an issue because LB reads
from /proc/cpuinfo.

I agree that the LB algorithm would seem to need some improving, both the
default automatic algorithm (possibly consider factoring in processor
generation i.e. Pentium 4 vs. Core 2, total RAM, or even current load and
available RAM) and the ability to manually tweak. 

William Yang

> -----Original Message-----
> From: sunray-users-bounces at filibeto.org [mailto:sunray-users-
> bounces at filibeto.org] On Behalf Of Sammy Atmadja
> Sent: Tuesday, March 16, 2010 3:35 AM
> To: SunRay-Users mailing list
> Subject: Re: [SunRay-Users] Failover group load balancing on servers with
> lots of cores
> 
> P.S.M.Swamiji wrote:
> > On 3/16/2010 3:04 AM, William Yang wrote:
> >> Are CPU frequency and count the only things the load balance algorithm
> >> considers?
> 
> It is consistent with the behaviour we're seeing:
> 
> server 1, 4 core opteron 1.8GHz, 6GiB RAM, doesn't support powersaving
> server 2, 4 core opteron 2.8GHz, 8GiB RAM, powernow/ondemand freq governor
> 
> In theory server 2 should be much more powerful, as it has more RAM and
> more
> powerful cpu's. But since /proc/cpuinfo shows it's clock freq as 1GHz
> because of
> the powernow cpu throttling, server 1 gets all the sessions, and runs out
> of RAM
> long before server 2 starts breaking sweat. As is mentioned on the sun-
> rays wiki
> somewhere, available RAM should be a much more important factor than CPU.
> I
> would much rather have somewhat slower cpu's and lots of free RAM than
> fast
> cpu's twiddling thumbs while waiting for swap.
> 
> 
> > It also considers number of sessions running/idle...
> >>   That concerns me a bit especially since with recent processors,
> >> increasing speed no longer necessarily means increasing performance.
> Can
> >> the LB algorithm be manually tweaked if necessary?
> 
> How about an external command (shell script?) that computes the
> "desirability"
> of the server in question? (feature request)
> 
> Sammy Atmadja
> This is a message from the E-MAIL server of Transtrend B.V.
> 
> The information contained in this communication is confidential and
> intended solely for the use of the individual or entity to whom it is
> addressed. You should not copy, disclose, or distribute this communication
> without the authority of Transtrend B.V. Transtrend B.V. is neither liable
> for the proper and complete transmission of the information contained in
> this communication nor for any delay in its receipt. Transtrend B.V. does
> not guarantee that the integrity of this communication has been maintained
> nor that the communication is free of viruses, interceptions or
> interference.
> 
> If you are not the intended recipient of this communication please return
> the communication to the sender and delete and destroy all copies.
> 
> De informatie verzonden in dit e-mailbericht is vertrouwelijk en
> uitsluitend bestemd voor de geadresseerde. Openbaarmaking,
> vermenigvuldiging, verspreiding en/of verstrekking van deze informatie aan
> derden is, behoudens voorafgaande schriftelijke toestemming van Transtrend
> B.V.  niet toegestaan. Transtrend B.V. staat niet in voor de juiste en
> volledige overbrenging van de inhoud van een verzonden e-mailbericht, noch
> voor tijdige ontvangst daarvan. Transtrend B.V. kan niet garanderen dat
> een
> verzonden e-mailbericht vrij is van virussen, noch dat e-mailberichten
> worden overgebracht zonder inbreuk of tussenkomst van onbevoegde derden.
> 
> Indien bovenstaand e-mailbericht niet aan u is gericht, verzoeken wij u.
> vriendelijk doch dringend het e-mailbericht te retourneren aan de afzender
> en het origineel en eventuele kopieen te verwijderen en te vernietigen.
> _______________________________________________
> SunRay-Users mailing list
> SunRay-Users at filibeto.org
> http://www.filibeto.org/mailman/listinfo/sunray-users



More information about the SunRay-Users mailing list