[Openstack] Ceph vs swift

Remo at italy1.com Remo at italy1.com
Tue Jun 3 13:53:49 UTC 2014


 

Grazie Vincenzo, 

I am very familiar on those docs / links below.


Ciao 

On 2014-06-03 06:32, Vincenzo Pii wrote: 

> Hi Remo, 
> 
> We
are doing a performance evaluation study on Ceph vs Swift for small
storage clusters. 
> The results should be published soon, so if the use
case is of interest to you you will have some material to analyze :). 
>

> Concerning the partition power, I think this article [1] (which is a
bit old btw) offers a good explanation of what the partitions are in
Swift and how they are used. 
> The swift deployment guide [2] also
provides some indications on how to determine the number of partitions
for the ring.
> 
> [1]:
https://julien.danjou.info/blog/2012/openstack-swift-consistency-analysis
[5] 
> [2]:
http://docs.openstack.org/developer/swift/deployment_guide.html#preparing-the-ring
[6] 
> 
> 2014-06-02 20:52 GMT+02:00 Chuck Thier <cthier at gmail.com>:
>

>> Hi Remo, 
>> 
>> I have heard that there has been some work to
integrate Hadoop with Swift, but know very little about it. Integration
with MS exchange, but could be an interesting use case. 
>> 
>>
Partitions can be thought of as virtual buckets that objects are
assigned to. They are an abstract concept and don't have any inherent
size. Buckets then are assigned to storage devices. The partition power
just determines how many partitions are distributed across the storage
devices. 
>> 
>> One of my coworkers wrote the following tool to help
calculate the partition power: http://rackerlabs.github.io/swift-ppc/
[4] 
>> 
>> -- 
>> Chuck 
>> 
>> On Fri, May 30, 2014 at 1:03 AM, Remo
Mattei <remo at italy1.com> wrote:
>> 
>>> Thanks Chuck 
>>> This is great
I want to use an object store which allows me to work well with Hadoop
and if possible with MS exchange IIS etc. If you do have tips on this
that will be great. I also would love your point of partition power in
how you describe it. I have my own ideas on it so I m curious to see how
to get an idea on object calculation on a partition. One of the question
I have always been asked was if the partition is sizex but the object is
sizez which is bigger how is that handle? 
>>> 
>>> Thanks for you
sharing I m sure everyone will take a good lesson. 
>>> 
>>> Ciao 
>>>

>>> Inviato da iPhone () 
>>> 
>>> Il giorno May 29, 2014, alle ore
22:39, Chuck Thier <cthier at gmail.com> ha scritto:
>>> 
>>>> Hello Remo,

>>>> 
>>>> That is quite an open ended question :) If you could share a
bit more about your use case, then it would be easier to provide more
detailed information, but I'll try to cover some of the basics. 
>>>>

>>>> First, a disclaimer. I am one of the original Openstack Swift
developers, so I *may* be a bit biased. 
>>>> 
>>>> Both Swift and Ceph
are capable object storage systems. Swift focuses purely on object
storage, while Ceph provides object, block and filesystem storage. This
leads to, what I believe is, the biggest fundamental difference between
Swift and Ceph. Since Ceph also provides block and filesystem storage,
it chooses consistency and partition tolerance over availability. Swift
was designed to be eventually consistent (similar to S3) favoring
availability and partition tolerance over consistency. If you are not
familiar with the CAP theorem, then I suggest starting with the
Wikipedia article about it [1] (oh and btw, all the images that you see
on Wikipedia are stored in a Swift cluster ;)) 
>>>> 
>>>> I can't speak
directly to the performance characteristics of Ceph, but I gave a talk
[2] at the Hong Kong summit where I go into quite a bit of detail about
how well swift has performed at Rackspace over the past 5 years, and how
far we have been able to scale it. Openstack Swift is also in use by a
substantial number of other companies and institutions all over the
world. 
>>>> 
>>>> All of that said, you can't trust anything you read
on the internet, or even me. If you are interested in running either
Swift or Ceph, then I would recommend evaluating your use case with both
systems. 
>>>> 
>>>> -- 
>>>> Chuck 
>>>> 
>>>> [1]
http://en.wikipedia.org/wiki/CAP_theorem [2] 
>>>> [2]
https://www.youtube.com/watch?v=YxnFUQHIwNs [3] 
>>>> 
>>>> On Thu, May
29, 2014 at 9:34 PM, Remo Mattei <remo at italy1.com> wrote:
>>>> 
>>>>> Hi
all does anyone have done any testing or comparison between swift and
ceph I want to get some others people prospectives.
>>>>> 
>>>>>
Thanks.
>>>>> 
>>>>> Inviato da iPhone ()
>>>>>
_______________________________________________
>>>>> Mailing list:
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack [1]
>>>>>
Post to : openstack at lists.openstack.org
>>>>> Unsubscribe :
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack [1]
>> 
>>
_______________________________________________
>> Mailing list:
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack [1]
>>
Post to : openstack at lists.openstack.org
>> Unsubscribe :
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack [1]
> 
>
-- 
> 
> Vincenzo Pii Researcher, InIT Cloud Computing Lab
> Zurich
University of Applied Sciences (ZHAW)
> http://www.cloudcomp.ch/ [7]
!DSPAM:1,538dcfe0103159641896319!

 

Links:
------
[1]
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
[2]
http://en.wikipedia.org/wiki/CAP_theorem
[3]
https://www.youtube.com/watch?v=YxnFUQHIwNs
[4]
http://rackerlabs.github.io/swift-ppc/
[5]
https://julien.danjou.info/blog/2012/openstack-swift-consistency-analysis
[6]
http://docs.openstack.org/developer/swift/deployment_guide.html#preparing-the-ring
[7]
http://www.cloudcomp.ch/
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20140603/2d947f9e/attachment.html>


More information about the Openstack mailing list