Re: Ceph Production Environment Setup and Configurations?

[Mark Nelson <mark.nelson@inktank.com>]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=GB2312, Size: 3024 bytes --]

On 01/29/2013 05:34 AM, Joao Eduardo Luis wrote:
> On 01/29/2013 11:25 AM, Chen, Xiaoxi wrote:
>> [The following views only behalf of myself, not relate with Intel..]
>> Looking forward for the performance data on Atom.
>> Atom perform badly in Swift, but since Ceph is slightly efficient than Swift, it must be better.
>> I have some concern about weather Atom can support such high throughput( you have 16 disks, assuming 50MB/s per disk, you would like to have 50*16 *2(include journal write)=1600MB/s ,together with corresponding network throughput, say 10GbE. The total IO throughput seems too high for Atom. Baidu (www.baidu.com) use Arm based storage node(not using ceph,but their own DFS), a node with a quad-core Arm together with 4 disks, and a 2U box can put up to 6 nodes, the 6 nodes share a 10Gb NIC. Although Atom is different with Arm, but you can take Baidu as a reference.
>>
>>
>> -----Original Message-----
>> From: ceph-devel-owner@vger.kernel.org [mailto:ceph-devel-owner@vger.kernel.org] On Behalf Of Gandalf Corvotempesta
>> Sent: 2013Äê1ÔÂ29ÈÕ 17:25
>> To: femi anjorin
>> Cc: ceph-devel@vger.kernel.org; Ross Turk
>> Subject: Re: Ceph Production Environment Setup and Configurations?
>>
>> 2013/1/29 femi anjorin <femi.anjorin@gmail.com>:
>>> CPU - Intel(R) Atom(TM) CPU D525   @ 1.80GHz
>>
>> Atom? For which kind of role ?
>> --
>> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in the body of a message to majordomo@vger.kernel.org More majordomo info at  http://vger.kernel.org/majordomo-info.html
>> N‹§²æìr¸›yúèšØb²X¬¶Ç§vØ^\x7f\x7f)Þº{.n\x7f\x7f+‰·œz˜]z÷¥Š{ay\x7f\x7f\x1dʇڙ\x7f\x7f,j\a­¢f£¢·hš‹àz\x7f\x7f\x1e®w¥¢\x7f\x7f\f¢·¦j:+v‰¨ŠwèjØm¶Ÿ\x7f\x7f\x7f\x7f\a«‘êçzZ+ƒùšŽŠÝ¢j"ú!tml=
>>
> 
> 
> FWIW, Wido had been testing Ceph on Atoms a while back.  I don't know
> what conclusions he reached, or even if he has already reached any
> conclusions at all.  I, for one, would be really interested in knowing
> how well (or how poorly) Atoms deal with Ceph.
> 
> Maybe Wido can share some thoughts on this one? :-)
> 
>    -Joao
> --
> To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Just FYI, I don't expect you'll be able to get anywhere close to 10GbE
performance on the Atom.  You'll almost certainly be CPU bound long
before that.  I actually would be curious how much throughput you could
push to those disks just doing straight fio tests and how much CPU
overhead you'd see.  I imagine you could max out the CPU without even
having Ceph involved.

I'm guessing you probably are going to be limited to about 0.75-1 OSD
per atom core, and that's assuming that your network and SAS controllers
aren't CPU hogs.

Mark
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html