You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@cassandra.apache.org by Mark Greene <gr...@gmail.com> on 2010/03/25 16:10:38 UTC

Separate disks with cloud deployment

The FAQ page makes mention of using separate disks for the commit log and
data directory. How would one go about achieving this in a cloud deployment
such as Rackspace cloud servers or EC2 EBS? Or is it just preferred to use
dedicated hardware to get the optimal performance?

Thanks In Advance!

Best,
Mark

Re: Separate disks with cloud deployment

Posted by Jonathan Ellis <jb...@gmail.com>.
If you have enough data or insert volume that you can reasonably use
dedicated hardware, you should probably use that.
(http://spyced.blogspot.com/2010/03/why-your-data-may-not-belong-in-cloud.html)

If you don't, then having CL + data on the same volume isn't going to
hurt nearly as much as sharing the machine with a bunch of other
users, so I wouldn't worry too much.

On Thu, Mar 25, 2010 at 10:10 AM, Mark Greene <gr...@gmail.com> wrote:
> The FAQ page makes mention of using separate disks for the commit log and
> data directory. How would one go about achieving this in a cloud deployment
> such as Rackspace cloud servers or EC2 EBS? Or is it just preferred to use
> dedicated hardware to get the optimal performance?
> Thanks In Advance!
> Best,
> Mark

Re: Separate disks with cloud deployment

Posted by Ethan Rowe <et...@endpoint.com>.
On 03/25/2010 11:18 AM, Ethan Rowe wrote:
> [snip]
> I'll defer to the Rackspace folks regarding Rackspace Cloud; it has 
> been I/O on average since you're dealing with a real, local disk.  But 
> I don't know about getting a second disk in that environment, though.

That should have said "better I/O on average".

-- 
Ethan Rowe
End Point Corporation
ethan@endpoint.com


Re: Separate disks with cloud deployment

Posted by Ethan Rowe <et...@endpoint.com>.
On 03/25/2010 11:10 AM, Mark Greene wrote:
> The FAQ page makes mention of using separate disks for the commit log 
> and data directory. How would one go about achieving this in a cloud 
> deployment such as Rackspace cloud servers or EC2 EBS? Or is it 
> just preferred to use dedicated hardware to get the optimal performance?

With EC2, you can mount more than one EBS device on a given server, so 
it's not a big deal.  Mount one volume for logs, another volume for data.

Additionally, we've found some benefit from running (for Postgres) RAID0 
arrays on EBS; you get better I/O throughput.

I'll defer to the Rackspace folks regarding Rackspace Cloud; it has been 
I/O on average since you're dealing with a real, local disk.  But I 
don't know about getting a second disk in that environment, though.

-- 
Ethan Rowe
End Point Corporation
ethan@endpoint.com