You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Lee Parker <le...@socialagency.com> on 2010/07/13 04:00:31 UTC
concurrent reads
Has anyone experimented with different settings for concurrent reads? I
have set our servers to 4 ( 2 per processor core ). I have noticed that
occasionally, our pending reads will get backed up and our servers don't
appear to be under too much load. In fact, most of the load appears to be
from GC. Is 3 per processor core too much? Does it matter if it is an AMC
vs Intel processor? How does processor clock speed or cache play into this
setting?
Lee Parker
Re: concurrent reads
Posted by Lee Parker <le...@socialagency.com>.
The iostat numbers are rather low as is cpu utilization. We have a couple
of nightly jobs which do a lot of reads in a short amount of time. That is
when the pending reads was climbing. I'm going to bump up the number and
see how things run.
Lee Parker
On Tue, Jul 13, 2010 at 6:18 AM, Schubert Zhang <zs...@gmail.com> wrote:
> For read, the bottleneck is usually the disk.
> Use iostat to check the utility of your disks.
>
>
>
>
> On Tue, Jul 13, 2010 at 2:07 PM, Peter Schuller <
> peter.schuller@infidyne.com> wrote:
>
>> > Has anyone experimented with different settings for concurrent reads? I
>> > have set our servers to 4 ( 2 per processor core ). I have noticed that
>> > occasionally, our pending reads will get backed up and our servers don't
>> > appear to be under too much load. In fact, most of the load appears to
>> be
>> > from GC. Is 3 per processor core too much? Does it matter if it is an
>> AMC
>> > vs Intel processor? How does processor clock speed or cache play into
>> this
>> > setting?
>>
>> Increase concurrency until you're able to either saturate CPU or
>> saturate the disk subsystem. Note that the more disks you have, the
>> more concurrency you'll need to fully utilize them.
>>
>> --
>> / Peter Schuller
>>
>
>
Re: concurrent reads
Posted by Schubert Zhang <zs...@gmail.com>.
For read, the bottleneck is usually the disk.
Use iostat to check the utility of your disks.
On Tue, Jul 13, 2010 at 2:07 PM, Peter Schuller <peter.schuller@infidyne.com
> wrote:
> > Has anyone experimented with different settings for concurrent reads? I
> > have set our servers to 4 ( 2 per processor core ). I have noticed that
> > occasionally, our pending reads will get backed up and our servers don't
> > appear to be under too much load. In fact, most of the load appears to
> be
> > from GC. Is 3 per processor core too much? Does it matter if it is an
> AMC
> > vs Intel processor? How does processor clock speed or cache play into
> this
> > setting?
>
> Increase concurrency until you're able to either saturate CPU or
> saturate the disk subsystem. Note that the more disks you have, the
> more concurrency you'll need to fully utilize them.
>
> --
> / Peter Schuller
>
Re: concurrent reads
Posted by Peter Schuller <pe...@infidyne.com>.
> Has anyone experimented with different settings for concurrent reads? I
> have set our servers to 4 ( 2 per processor core ). I have noticed that
> occasionally, our pending reads will get backed up and our servers don't
> appear to be under too much load. In fact, most of the load appears to be
> from GC. Is 3 per processor core too much? Does it matter if it is an AMC
> vs Intel processor? How does processor clock speed or cache play into this
> setting?
Increase concurrency until you're able to either saturate CPU or
saturate the disk subsystem. Note that the more disks you have, the
more concurrency you'll need to fully utilize them.
--
/ Peter Schuller
Re: concurrent reads
Posted by Jonathan Ellis <jb...@gmail.com>.
if you're not sure where your bottleneck is, you aren't hitting it
hard enough :)
On Mon, Jul 12, 2010 at 9:00 PM, Lee Parker <le...@socialagency.com> wrote:
> Has anyone experimented with different settings for concurrent reads? I
> have set our servers to 4 ( 2 per processor core ). I have noticed that
> occasionally, our pending reads will get backed up and our servers don't
> appear to be under too much load. In fact, most of the load appears to be
> from GC. Is 3 per processor core too much? Does it matter if it is an AMC
> vs Intel processor? How does processor clock speed or cache play into this
> setting?
>
> Lee Parker
--
Jonathan Ellis
Project Chair, Apache Cassandra
co-founder of Riptano, the source for professional Cassandra support
http://riptano.com