[Gluster-users] AFR performance killer

[Marko]

It's physically 4 SATA disks, connected via 3ware RAID controller
configured as RAID 1+0.
On this RAID10 I have one big LVM volume from which I allocated Logical
Volumes to Xen guests.

I must point out that this is my staging environment. I can't put it in
production yet. I must get some satisfactory
results first.

In production GlusterFS will reside on a bunch of servers with 2 HDDs
with software mirror.

Regards,
Marko

Shehjar Tikoo wrote:
> Marko wrote:
>> All Xen domains share same physical volume but have theier uniqe
>> logical volume.
> Please be a bit more specific about the volume. Is it a single large
> disk or a bunch of RAIDx disks?
>
> Thanks
> Shehjar
>
>> So, yes I'd expect performance degradation becouse of that (that is
>> why test DHT2 performed better than DHT4). But I cant expect
>> performance drop when I use a performance "booster".
>>
>> Regards, Marko
>>
>> Shehjar Tikoo wrote:
>>> Marko wrote:
>>>> Hi Shehjar,
>>>>
>>>> Answers: 1.) Filesystem is default ext3 mounted with
>>>> "defaults,noatime"
>>>
>>> I see that each server, i.e. iscsi1 and iscsi2, has /srv/gluster as
>>> the backend in their corresponding xen domains, but did the xen
>>> domains also share the same disks/storage device?
>>>
>>> Thanks Shehjar
>>>
>>>> 2.) I think it how it was becouse I always put io-threads as
>>>> top(or near top) translator.  Iin TEST5 "iot" can be ignored. 3.)
>>>>  Server(brick) iscsi2 was also client on which I was running the
>>>> test and it was also a Xen guest.
>>>>
>>>> One more thing I'd like to point out. In documentation I miss
>>>> information of best practices which translators should be put
>>>> before/after other translators. Example: should one put
>>>> io-threads before or after io-cache? What does it mean if you
>>>> configure either way.
>>>>
>>>> Regards, Marko
>>>>
>>>> Shehjar Tikoo wrote:
>>>>> Marko wrote:
>>>>>> this document is made for my personal reference so it's a
>>>>>> little raw.
>>>>>>
>>>>>>
>>>>>
>>>>> A few questions:
>>>>>
>>>>> 1. What was the on-disk file system and the disk organization
>>>>> under the file system?
>>>>>
>>>>> 2. In the TEST5 volfile, the volume iot is dangling. Is that
>>>>> how it actually was during the test or was it interposed
>>>>> between volume wb and volume AFR?
>>>>>
>>>>>>> On Wed, Apr 8, 2009 at 6:18 PM, Marko <gluster at sopca.com
>>>>>>> <mailto:gluster at sopca.com>> wrote:
>>>>>>>
>>>>>>> Hello,
>>>>>>>
>>>>>>> To clearify: * im testing with gluisterfs-2.0.0rc7 * all
>>>>>>> bricks are on same physical server(Xen guests). It's a
>>>>>>> testing environment.
>>>>>
>>>>> Were the client side tests( i.e., both the test program like
>>>>> make-many-files and the corresponding glusterfs client..) also
>>>>>  running in a Xen guest?
>>>>>
>>>>> Thanks Shehjar
>>>>>
>>>>>
>>>>
>>>>
>>>
>>>
>>
>>
>
>