questions about radosgw

questions about radosgw

[huang jun]

hi,all
I'am using the s3+radosgw, there are some questions confused me very much:
first) An object's size is 400MB, then the  whole object will stored
in OSDs as one big single object, but not striped into 4MB objects. so
how can we got workload banlance if we want to store big objects and
small ones?
second) Can we change the number of pgs in a pool that created by S3
clients or just using rados commands?

thanks

-- 
huangjun

Re: questions about radosgw

[Wido den Hollander]

Hi,

On 01/05/2012 12:48 PM, huang jun wrote:
> hi,all
> I'am using the s3+radosgw, there are some questions confused me very much:
> first) An object's size is 400MB, then the  whole object will stored
> in OSDs as one big single object, but not striped into 4MB objects. so
> how can we got workload banlance if we want to store big objects and
> small ones?

That is correct, RADOS nor the RADOS gateway stripes objects.

You will get workload balance if you have a (almost) even spread of 
workload over your different objects.

The dev's might shed some more light on this.

> second) Can we change the number of pgs in a pool that created by S3
> clients or just using rados commands?

As far as I know you can't do this through the S3 clients but only with 
the rados commands. I also would find it rather weird if you could do 
this as a client.

Wido

>
> thanks
>

Re: questions about radosgw

[huang jun]

hi,wido
I have an idea, why not use radosgw+fs to balance the workload?
but i can not successfully deploy it.
I really want to know whether radosgw+fs works?
If so, how about the performance between radosgw+fs and radosgw+rados?

thanks!

2012/1/5 Wido den Hollander <wido@widodh.nl>:
> Hi,
>
>
> On 01/05/2012 12:48 PM, huang jun wrote:
>>
>> hi,all
>> I'am using the s3+radosgw, there are some questions confused me very much:
>> first) An object's size is 400MB, then the  whole object will stored
>> in OSDs as one big single object, but not striped into 4MB objects. so
>> how can we got workload banlance if we want to store big objects and
>> small ones?
>
>
> That is correct, RADOS nor the RADOS gateway stripes objects.
>
> You will get workload balance if you have a (almost) even spread of workload
> over your different objects.
>
> The dev's might shed some more light on this.
>
>
>> second) Can we change the number of pgs in a pool that created by S3
>> clients or just using rados commands?
>
>
> As far as I know you can't do this through the S3 clients but only with the
> rados commands. I also would find it rather weird if you could do this as a
> client.
>
> Wido
>
>>
>> thanks
>>
>



-- 
huangjun
--
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

Re: questions about radosgw

[Yehuda Sadeh Weinraub]

On Fri, Jan 6, 2012 at 7:10 PM, huang jun <hjwsm1989@gmail.com> wrote:
> I have an idea, why not use radosgw+fs to balance the workload?
> but i can not successfully deploy it.
> I really want to know whether radosgw+fs works?
> If so, how about the performance between radosgw+fs and radosgw+rados?
>

At this point radosgw cannot use arbitrary file system as a backend.
This requires a restructuring of the code, though we do have that in
mind. The current 'fs' backend for radosgw is obsolete and wouldn't
work anyway as the internal interfaces evolved quite a bit since it
had been written, mainly to take advantage of rados capabilities that
a regular filesystem cannot provide.
The plan is to create a new devel library (similar to librados and
librbd) that will provide a filesystem-like api. Having radosgw using
this library (instead of the current internal access library) will
help enabling a fs backend. But, as I said, there's still some work
ahead of us.

Yehuda

Re: questions about radosgw

[huang jun]

2012/1/7 Yehuda Sadeh Weinraub <yehudasa@gmail.com>:
> On Fri, Jan 6, 2012 at 7:10 PM, huang jun <hjwsm1989@gmail.com> wrote:
>> I have an idea, why not use radosgw+fs to balance the workload?
>> but i can not successfully deploy it.
>> I really want to know whether radosgw+fs works?
>> If so, how about the performance between radosgw+fs and radosgw+rados?
>>
>
> At this point radosgw cannot use arbitrary file system as a backend.
> This requires a restructuring of the code, though we do have that in
> mind. The current 'fs' backend for radosgw is obsolete and wouldn't
> work anyway as the internal interfaces evolved quite a bit since it
> had been written, mainly to take advantage of rados capabilities that
> a regular filesystem cannot provide.
> The plan is to create a new devel library (similar to librados and
> librbd) that will provide a filesystem-like api. Having radosgw using
> this library (instead of the current internal access library) will
> help enabling a fs backend. But, as I said, there's still some work
> ahead of us.
>
> Yehuda
hi, thanks for you reply
so the workload balance will be a hidden trouble using radosgw?do you
have any ideas to promote this?
and if we put an object in atmotic way: put it to a temp object and
after the write finished , send a clone op, does this have any effect
on IO thourghput?




-- 
huangjun
ChuYunHanZhi WuHan Network Storage System Ltd
WuHan GuangGu Road