Re: Raid0 with btrfs

[Leonidas Spyropoulos <artafinde@gmail.com>]

On Sat, Aug 7, 2010 at 1:03 AM, Sebastian 'gonX' Jensen
<gonx@overclocked.net> wrote:
> On 7 August 2010 00:24, Leonidas Spyropoulos <artafinde@gmail.com> wr=
ote:
>> On Fri, Aug 6, 2010 at 7:32 PM, Sebastian 'gonX' Jensen
>> <gonx@overclocked.net> wrote:
>>> On 6 August 2010 20:23, Leonidas Spyropoulos <artafinde@gmail.com> =
wrote:
>>>> Do I have to change the partition ID flag of each partition?
>>>> Currently is set to fd (Linux Raid autodetect) for used from mdadm
>>>> mkfs.btrfs supports that or needs to be 83 (Linux) ?
>>>
>>> FD is for mdraid integrated into the Linux kernel. I have mine at 8=
3.
>>> It won't hurt to have them at FD, but the kernel will spend extra t=
ime
>>> as it probes the devices on boot, causing a slight slowdown.
>>
>> Ok done them id 83 and used 3 devices eventually
>> Using raid0 for data and metadata
>> # mkfs.btrfs -m raid0 -d raid0 /dev/sdc1 /dev/sdd2 /dev/sde2
>>
>> 2 SATA and 1 ATA drive
>> I thought that ATA will bottleneck the other 2 drives but seems like=
 I gain
>> something from it.
>> Using iozone for benchmark:
>> # iozone -s 8g -r 1024 -i 0 -i 1
>> with 2 SATA devices and then 3 devices (SATA + ATA):
>> =A0 =A0 =A0 =A0KB =A0reclen =A0 =A0 =A0write =A0 rewrite =A0 =A0 =A0=
 read =A0 =A0 reread
>> 8388608 =A0 =A01024 =A0134869 =A0139607 =A0 229146 =A0 228800
>> 8388608 =A0 =A01024 =A0135151 =A0139050 =A0 233461 =A0 235929
>>
>> The above is with -o compress option enabled and my cpu topped up on
>> 100% cpu (both cores) while test and copy huge data.
>> Is it possible I am bottlenecked by my cpu speed?
>> AMD Opteron 165 @ 2700 Mhz
>>
>>>>
>>>> On Thu, Aug 5, 2010 at 3:40 PM, Hubert Kario <hka@qbs.com.pl> wrot=
e:
>>>>> On Thursday 05 August 2010 16:15:22 Leonidas Spyropoulos wrote:
>>>>>> Hi all,
>>>>>>
>>>>>> I want to make a btrfs raid0 on 2 partitions of my pc.
>>>>>> Until now I am using the mdadm tools to make a software raid of =
the 2
>>>>>> partitions /dev/sde2, /dev/sdd2
>>>>>> and then mkfs.etx4 the newly created /dev/md0 device.
>>>>>> From performance point of view is it better to keep the configur=
ation of
>>>>>> mdadm and just format the /dev/md0 device as btrfs OR
>>>>>> delete the raid device and format the 2 partitions /dev/sde2 /de=
v/sdd2
>>>>>> as a btrfs with 2 devices?
>>>>>> mkfs.btrfs /dev/sde2 /dev/sdd2
>>>>>
>>>>> Btrfs already supports metadata mirroring when the data is stripe=
d. What this
>>>>> means, is while the performance should be more-or-less identical =
to MD RAID0
>>>>> (if it isn't it's a bug), your data is a bit more secure as the m=
etadata
>>>>> describing it resides on both drives. Later on it will be possibl=
e to selct
>>>>> which directories/files should have what level of redundancy. Thi=
s will allow
>>>>> to have ~/work RAID1-ed and ~/videos RAID0-ed while keeping both =
directories
>>>>> on the same partition and filesystem.
>>>>>
>>>>>> On a sidenote:
>>>>>> If I decide to go for raid5 which is not supported currently fro=
m mkfs
>>>>>> I have to use the mdadm tool anyway, right?
>>>>>
>>>>> yes, RAID5 code is not in trunk yet.
>>>>>
>>>>> --
>>>>> Hubert Kario
>>>>> QBS - Quality Business Software
>>>>> 02-656 Warszawa, ul. Ksawer=F3w 30/85
>>>>> tel. +48 (22) 646-61-51, 646-74-24
>>>>> www.qbs.com.pl
>>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Caution: breathing may be hazardous to your health.
>>>> --
>>>> To unsubscribe from this list: send the line "unsubscribe linux-bt=
rfs" in
>>>> the body of a message to majordomo@vger.kernel.org
>>>> More majordomo info at =A0http://vger.kernel.org/majordomo-info.ht=
ml
>>>>
>>>
>>> Regards,
>>> Sebastian J.
>>>
>>> PS. Please try to bottom-post rather than top-post. Here's a link I
>>> can advise reading for a clarification on why bottom posting is
>>> essential: http://www.caliburn.nl/topposting.html
>>>
>> Thanks for the heads up about bottom-posting.
>>
>>
>> --
>> Caution: breathing may be hazardous to your health.
>>
>
> It seems odd that you are reaching 100% CPU usage. Maybe it's because
> I am on an Intel processor with the integrated crc32c module, but wit=
h
> metadata and data striping on 2 decent desktop drives (60 MB/s in avg
> each on large reads), I don't go much anywhere near 10% CPU usage, an=
d
> I have compress-force enabled. I've never tried a stress test iozone,
> but if I can still remember it, I will try to run it once I get back
> from holidays. Sorry that it doesn't help your problem, but it seems
> like it's something else.

I really doubt my old Opteron has SSE 4.2 and as mentioned from other
users it makes a huge difference.
>
> I'm assuming you're running the nForce4 chipset. I don't recall it
> being there, but is there an AHCI option for S-ATA in the BIOS, rathe=
r
> than legacy or PATA mode, or something in the lines of that? That
> could in theory reduce CPU usage somewhat, but shouldn't really affec=
t
> anything before very high transfer speeds.
>
Yes it's an nForce4 chipset (DFI Expert)
And if I recall it's SATA+PATA mode enabled on mine now.
I can't be sure since I am working on it from ssh.

> And yes, you are bottlenecked if you're running at max CPU usage. I
> would try disabling the compress mount option if the above does not
> help.
After disabling the compress and done the test again it affected the wr=
ite speed
and my CPU wasn't topped up all the time something like 80-90%
results with iozone
        KB   reclen     write   rewrite       read     reread
8388608    1024  147736  147062   135427   134744

thanks for that
>
> Regards,
> Sebastian J.
>

The next step is remove the ATA drive and test again with and without
compress mount.


--=20
Caution: breathing may be hazardous to your health.
--
To unsubscribe from this list: send the line "unsubscribe linux-btrfs" =
in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html