I am using "sheep -n /mnt/sheep/0 -c zookeeper:IP1:2181,IP2:2181,IP3:2181"
to start sheep on each node. And in the question I was mostly talking about
reading speed. Writing speed was quite less than the native disk writing.

On my machine I am able to run 10 VMs comfortably , within 40 seconds when
I am using local disk. But when I try to boot these VM in 3 node- 3 replica
sheepdog cluster, CPU becomes the bottleneck after booting 3 VM
simultaneously.  Here is the graph for booing 5 VM simultaneously in
sheepdog cluster.


​

I want to run more number of VMs. Is there anything, that can be done in
terms of setup, number of node, Number of replicas etc ?


On Tue, Jun 23, 2015 at 5:51 PM, Hitoshi Mitake <[email protected]>
wrote:

> At Tue, 23 Jun 2015 11:57:53 +0530,
> ankit bhardwaj wrote:
> >
> > [1  <multipart/alternative (7bit)>]
> > [1.1  <text/plain; UTF-8 (7bit)>]
> > I am using 3 node cluster , with 3 copies of disk image. So, each node
> have
> > a full copy of disk image.
>
> Could you provide an option for your sheep processes?
>
> Thanks,
> Hitoshi
>
> >
> > On Tue, Jun 23, 2015 at 10:37 AM, Vasiliy Tolstov <[email protected]>
> > wrote:
> >
> > > 2015-06-23 1:40 GMT+03:00 ankit bhardwaj <[email protected]>:
> > > > I am new to Sheepdog. Though I have gained enough knowledge to
> deploy and
> > > > use sheepdog for qemu-kvm VMs. I tried to benchmark Sheepdog and
> found
> > > that:
> > > >
> > > > 1. Sheepdog's Maximum Read/Write Speed is around 70-80% compared to
> local
> > > > storage. That too when the condition are in favour of sheepdog.
> Example:
> > > > Full copy of data on same node and Reading or writing sequential Data
> > > etc.
> > > >
> > > > 2. When using VM disk image to boot VM from Sheepdog storage, some
> extra
> > > > system activities increase the load on the CPU from 30-40%.
> > > >
> > > > Can you please tell me,:
> > > > Why the speed is slow, even if all the data is on the same machine ?
> > > > What activity in sheepdog is eating up my processor cycles ?
> > > >
> > > >
> > > > Sorry for the loosely described experiments and results.
> > >
> > >
> > > I think we need command line for running sheepdog and how you format
> > > you cluster (erasure coded or replicas) ?
> > >
> > > --
> > > Vasiliy Tolstov,
> > > e-mail: [email protected]
> > >
> >
> >
> >
> > --
> > Ankit Bhardwaj
> > [1.2  <text/html; UTF-8 (quoted-printable)>]
> >
> > [2  <text/plain; us-ascii (7bit)>]
> > --
> > sheepdog mailing list
> > [email protected]
> > https://lists.wpkg.org/mailman/listinfo/sheepdog
>



-- 
Ankit Bhardwaj
-- 
sheepdog mailing list
[email protected]
https://lists.wpkg.org/mailman/listinfo/sheepdog

Reply via email to