massive 4k random read drop with next branch

massive 4k random read drop with next branch

[Stefan Priebe]

Hello list,

today i was testing next branch against bobtail using a testcluster. I 
upgraded this bobtail one to next branch.

I used QEMU 1.4.1 with librdb (for next branch i applied josh writeback 
patch). My Testsystem had 5 Hosts with one OSD each and set replication 
to 2.

All values where done with 8 jobs in parallel using fio in Qemu Guest.

bobtail:
rand 4k:  write: io=164316KB, bw=1783KB/s, iops=445, runt= 92136msec
rand 4k:  read : io=1117MB, bw=12710KB/s, iops=3177, runt= 90028msec
seq 4m:  write: io=8588MB, bw=97233KB/s, iops=23, runt= 90444msec
seq 4m:  read : io=83616MB, bw=951227KB/s, iops=232, runt= 90013msec

next branch:
   write: io=177236KB, bw=1963KB/s, iops=490, runt= 90284msec
   read : io=223628KB, bw=2408KB/s, iops=601, runt= 92875msec
   write: io=25936MB, bw=294443KB/s, iops=71, runt= 90199msec
   read : io=69856MB, bw=794585KB/s, iops=193, runt= 90025msec

Greets,
Stefan

Re: massive 4k random read drop with next branch

[Mark Nelson]

On 04/27/2013 03:13 PM, Stefan Priebe wrote:
> Hello list,
>
> today i was testing next branch against bobtail using a testcluster. I
> upgraded this bobtail one to next branch.
>
> I used QEMU 1.4.1 with librdb (for next branch i applied josh writeback
> patch). My Testsystem had 5 Hosts with one OSD each and set replication
> to 2.
>
> All values where done with 8 jobs in parallel using fio in Qemu Guest.
>
> bobtail:
> rand 4k:  write: io=164316KB, bw=1783KB/s, iops=445, runt= 92136msec
> rand 4k:  read : io=1117MB, bw=12710KB/s, iops=3177, runt= 90028msec
> seq 4m:  write: io=8588MB, bw=97233KB/s, iops=23, runt= 90444msec
> seq 4m:  read : io=83616MB, bw=951227KB/s, iops=232, runt= 90013msec
>
> next branch:
>    write: io=177236KB, bw=1963KB/s, iops=490, runt= 90284msec
>    read : io=223628KB, bw=2408KB/s, iops=601, runt= 92875msec
>    write: io=25936MB, bw=294443KB/s, iops=71, runt= 90199msec
>    read : io=69856MB, bw=794585KB/s, iops=193, runt= 90025msec

Was everything else the same between the tests?  That does seem unusual. 
  If it's easy for you to switch between bobtail and next, could you try 
some 4k rados bench read tests to see if it's also happening there?

>
> Greets,
> Stefan
>
> --
> 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: massive 4k random read drop with next branch

[Wido den Hollander]

On 04/27/2013 10:13 PM, Stefan Priebe wrote:
> Hello list,
>
> today i was testing next branch against bobtail using a testcluster. I
> upgraded this bobtail one to next branch.
>
> I used QEMU 1.4.1 with librdb (for next branch i applied josh writeback
> patch). My Testsystem had 5 Hosts with one OSD each and set replication
> to 2.
>
> All values where done with 8 jobs in parallel using fio in Qemu Guest.
>

I'm not seeing the same. Although my tests were with Bonnie++ I actually 
noticed a very big increase.

The biggest increase was write performance which went up a lot going 
from Bobtail to next, but with next VMs (Qemu 1.2.2) are a lot more 
snappier.

The benchmarks don't give me a huge difference in read performance, but 
my VMs feel a lot faster however.

Wido

> bobtail:
> rand 4k:  write: io=164316KB, bw=1783KB/s, iops=445, runt= 92136msec
> rand 4k:  read : io=1117MB, bw=12710KB/s, iops=3177, runt= 90028msec
> seq 4m:  write: io=8588MB, bw=97233KB/s, iops=23, runt= 90444msec
> seq 4m:  read : io=83616MB, bw=951227KB/s, iops=232, runt= 90013msec
>
> next branch:
>    write: io=177236KB, bw=1963KB/s, iops=490, runt= 90284msec
>    read : io=223628KB, bw=2408KB/s, iops=601, runt= 92875msec
>    write: io=25936MB, bw=294443KB/s, iops=71, runt= 90199msec
>    read : io=69856MB, bw=794585KB/s, iops=193, runt= 90025msec
>
> Greets,
> Stefan
>
> --
> 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


-- 
Wido den Hollander
42on B.V.

Phone: +31 (0)20 700 9902
Skype: contact42on

Re: massive 4k random read drop with next branch

[Stefan Priebe - Profihost AG]

Am 27.04.2013 um 22:45 schrieb Mark Nelson <mark.nelson@inktank.com>:

> On 04/27/2013 03:13 PM, Stefan Priebe wrote:
>> Hello list,
>> 
>> today i was testing next branch against bobtail using a testcluster. I
>> upgraded this bobtail one to next branch.
>> 
>> I used QEMU 1.4.1 with librdb (for next branch i applied josh writeback
>> patch). My Testsystem had 5 Hosts with one OSD each and set replication
>> to 2.
>> 
>> All values where done with 8 jobs in parallel using fio in Qemu Guest.
>> 
>> bobtail:
>> rand 4k:  write: io=164316KB, bw=1783KB/s, iops=445, runt= 92136msec
>> rand 4k:  read : io=1117MB, bw=12710KB/s, iops=3177, runt= 90028msec
>> seq 4m:  write: io=8588MB, bw=97233KB/s, iops=23, runt= 90444msec
>> seq 4m:  read : io=83616MB, bw=951227KB/s, iops=232, runt= 90013msec
>> 
>> next branch:
>>   write: io=177236KB, bw=1963KB/s, iops=490, runt= 90284msec
>>   read : io=223628KB, bw=2408KB/s, iops=601, runt= 92875msec
>>   write: io=25936MB, bw=294443KB/s, iops=71, runt= 90199msec
>>   read : io=69856MB, bw=794585KB/s, iops=193, runt= 90025msec
> 
> Was everything else the same between the tests?  That does seem unusual.  If it's easy for you to switch between bobtail and next, could you try some 4k rados bench read tests to see if it's also happening there?

I think I got it. Under bobtail the whole stuff was in cache on the osd clients. While under next I tested while they were freshly rebooted ;-( sorry guys.

Will retest.

Stefan
> 
>> 
>> Greets,
>> Stefan
>> 
>> --
>> 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: massive 4k random read drop with next branch

[Mark Nelson]

On 04/28/2013 07:21 AM, Stefan Priebe - Profihost AG wrote:
> Am 27.04.2013 um 22:45 schrieb Mark Nelson <mark.nelson@inktank.com>:
>
>> On 04/27/2013 03:13 PM, Stefan Priebe wrote:
>>> Hello list,
>>>
>>> today i was testing next branch against bobtail using a testcluster. I
>>> upgraded this bobtail one to next branch.
>>>
>>> I used QEMU 1.4.1 with librdb (for next branch i applied josh writeback
>>> patch). My Testsystem had 5 Hosts with one OSD each and set replication
>>> to 2.
>>>
>>> All values where done with 8 jobs in parallel using fio in Qemu Guest.
>>>
>>> bobtail:
>>> rand 4k:  write: io=164316KB, bw=1783KB/s, iops=445, runt= 92136msec
>>> rand 4k:  read : io=1117MB, bw=12710KB/s, iops=3177, runt= 90028msec
>>> seq 4m:  write: io=8588MB, bw=97233KB/s, iops=23, runt= 90444msec
>>> seq 4m:  read : io=83616MB, bw=951227KB/s, iops=232, runt= 90013msec
>>>
>>> next branch:
>>>    write: io=177236KB, bw=1963KB/s, iops=490, runt= 90284msec
>>>    read : io=223628KB, bw=2408KB/s, iops=601, runt= 92875msec
>>>    write: io=25936MB, bw=294443KB/s, iops=71, runt= 90199msec
>>>    read : io=69856MB, bw=794585KB/s, iops=193, runt= 90025msec
>>
>> Was everything else the same between the tests?  That does seem unusual.  If it's easy for you to switch between bobtail and next, could you try some 4k rados bench read tests to see if it's also happening there?
>
> I think I got it. Under bobtail the whole stuff was in cache on the osd clients. While under next I tested while they were freshly rebooted ;-( sorry guys.
>
> Will retest.

No worries!  Unless I'm explicitly testing from cache, I always do a 
sync and drop_caches on both the clients and servers before doing any tests.

>
> Stefan
>>
>>>
>>> Greets,
>>> Stefan
>>>
>>> --
>>> 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
>>