* Load Balancing Performance
@ 2003-03-11 9:58 chandrasekhar.nagaraj
2003-03-11 10:02 ` Jakob Oestergaard
0 siblings, 1 reply; 2+ messages in thread
From: chandrasekhar.nagaraj @ 2003-03-11 9:58 UTC (permalink / raw)
To: linux-kernel
Hi,
We have a device driver which is on top of the "sd" driver (i.e it forwards
the requests to the sd driver with various features implemented like load
balance,fail over etc).
Suppose we are having 4 paths to the same Logical Unit(LU),it is possible to
do load balance amongst these 4 paths.
Now if we want to send a data of 1000 buffer heads ,on each call of the make
request of our driver we select the appropriate path and pass it to sd
(this is how we achieve load balancing) and hence 250 buffer heads are sent
through each path.But the performance is not up to the mark.
But instead of path switching for every buffer head (on call of make
request), if we switch after 50 buffer heads(ie. after 50 times make
request is called) then it results in a better performance.
Is there any way to find out the optimum value so that the performance is
optimum.
The performance is not good only when load balancing is ON. Hence we need
performance improvement in the case of load balancing.
Or is there any other solution to this problem.
Thanks and Regards
Chandrasekhar
^ permalink raw reply [flat|nested] 2+ messages in thread
* Re: Load Balancing Performance
2003-03-11 9:58 Load Balancing Performance chandrasekhar.nagaraj
@ 2003-03-11 10:02 ` Jakob Oestergaard
0 siblings, 0 replies; 2+ messages in thread
From: Jakob Oestergaard @ 2003-03-11 10:02 UTC (permalink / raw)
To: chandrasekhar.nagaraj; +Cc: linux-kernel
On Tue, Mar 11, 2003 at 03:28:14PM +0530, chandrasekhar.nagaraj wrote:
> Hi,
>
...
> But instead of path switching for every buffer head (on call of make
> request), if we switch after 50 buffer heads(ie. after 50 times make
> request is called) then it results in a better performance.
> Is there any way to find out the optimum value so that the performance is
> optimum.
> The performance is not good only when load balancing is ON. Hence we need
> performance improvement in the case of load balancing.
> Or is there any other solution to this problem.
Make the request site tunable via. procfs.
Then create a (user space) utility which experiments with various
request sizes and finds the optimum. Run the tuning program once every
time you set up a new system or change hardware.
--
................................................................
: jakob@unthought.net : And I see the elder races, :
:.........................: putrid forms of man :
: Jakob Østergaard : See him rise and claim the earth, :
: OZ9ABN : his downfall is at hand. :
:.........................:............{Konkhra}...............:
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2003-03-11 9:51 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2003-03-11 9:58 Load Balancing Performance chandrasekhar.nagaraj
2003-03-11 10:02 ` Jakob Oestergaard
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.