DQ965GF poor RAID 5 write performance

Hi

Apologies if this is posted in the wrong group, if so could you point
me in the right direction.

I've just built a new home server; operating system is Server 2003 R2
standard. Motherboard is an Intel DQ965GF with a Core2Duo 1.86 and 2GB
DDR2 Kingston RAM.

I've been doing some testing before posting so hopefully you'll be able
to help me out.

In the system I have 3x500GB Seagate SATA drives and 1x320GB Maxtor
SATA drive. Initially I created a RAID 5 with the 3x500's. I
partitioned that into a 30GB for the system and that left somewhere
over 900 for the data.

Once it was up and running I updated the drivers and then tested the
drive performance. I used YAPT. Read performance on both partitions
was as I had hoped, it was around 130meg/sec. Write performance on the
data partition was 90meg/sec. However write performance on the system
partition was only 10meg/sec. I've checked the net and can't find
anything and I've confirmed write cache is on.

The next thing I did was to brake the RAID and have the drives as stand
alone. I installed Server on the 320GB maxtor and then tested
performance. The 3 Seagates all returned consistent 72meg/sec read and
write. The maxtor returned 62meg/sec read and write.

I've now recreated the RAID 5 and installed Server again but all as 1
big partition. Read performance is back up to 130's,eg/sec but write
is still 10meg/sec.

It's gone 1AM in the morning now but I might try XP rather than server
tomorrow but I'm sure that will be a waste of time.

Help

Many Thanks

Dan

wrote:
> Hi
>
> Apologies if this is posted in the wrong group, if so could you point
> me in the right direction.
>
> I've just built a new home server; operating system is Server 2003 R2
> standard. Motherboard is an Intel DQ965GF with a Core2Duo 1.86 and 2GB
> DDR2 Kingston RAM.

It's typical for RAID5 write performance. Also remember that this is a
software RAID5 rather than a hardware RAID5, which makes it even worse.

Yousuf Khan

On Dec 21, 4:04 pm, Yousuf Khan <bbb...@yahoo.com> wrote:
> mrdanielmor...@gmail.com wrote:
> > Hi
>
> > Apologies if this is posted in the wrong group, if so could you point
> > me in the right direction.
>
> > I've just built a new home server; operating system is Server 2003 R2
> > standard. Motherboard is an Intel DQ965GF with a Core2Duo 1.86 and 2GB
> > DDR2 Kingston RAM.It's typical for RAID5 write performance. Also remember that this is a
> software RAID5 rather than a hardware RAID5, which makes it even worse.
>
> Yousuf Khan

Software RAID? are you sure

wrote:

> On Dec 21, 4:04 pm, Yousuf Khan <bbb...@yahoo.com> wrote:
>> mrdanielmor...@gmail.com wrote:
>>> Hi
>>> Apologies if this is posted in the wrong group, if so could you point
>>> me in the right direction.
>>> I've just built a new home server; operating system is Server 2003 R2
>>> standard. Motherboard is an Intel DQ965GF with a Core2Duo 1.86 and 2GB
>>> DDR2 Kingston RAM.It's typical for RAID5 write performance. Also remember that this is a
>> software RAID5 rather than a hardware RAID5, which makes it even worse.

>> Yousuf Khan

> Software RAID? are you sure

Yup, I'm afraid so. You can't call it hardware RAID unless you have a
separate little processor doing all of the parity calculations for you.
All motherboard RAIDs are a big rip-off, because of the parity
calculations are being done by the main processor.

Yousuf Khan

--
There is no failure, only delayed success

* :

> I've just built a new home server; operating system is Server 2003 R2
> standard. Motherboard is an Intel DQ965GF with a Core2Duo 1.86 and 2GB
> DDR2 Kingston RAM.
>
> I've been doing some testing before posting so hopefully you'll be able
> to help me out.
>
> In the system I have 3x500GB Seagate SATA drives and 1x320GB Maxtor
> SATA drive. Initially I created a RAID 5 with the 3x500's. I
> partitioned that into a 30GB for the system and that left somewhere
> over 900 for the data.
>
> Once it was up and running I updated the drivers and then tested the
> drive performance. I used YAPT. Read performance on both partitions
> was as I had hoped, it was around 130meg/sec. Write performance on the
> data partition was 90meg/sec. However write performance on the system
> partition was only 10meg/sec. I've checked the net and can't find
> anything and I've confirmed write cache is on.
>
> The next thing I did was to brake the RAID and have the drives as stand
> alone. I installed Server on the 320GB maxtor and then tested
> performance. The 3 Seagates all returned consistent 72meg/sec read and
> write. The maxtor returned 62meg/sec read and write.
>
> I've now recreated the RAID 5 and installed Server again but all as 1
> big partition. Read performance is back up to 130's,eg/sec but write
> is still 10meg/sec.
>
> It's gone 1AM in the morning now but I might try XP rather than server
> tomorrow but I'm sure that will be a waste of time.

Yousuf already told you that you have a software RAID, so don't expect
miracles. But usually a software RAID is quite fast on todays processors
if the load isn't too big (and that shouldn't be the case on a home server).

As to your problem, I think the results you get are bogus. 130MB/s with
three 7200rpm SATA drives in RAID5 is a bit too optimistic. I don't know
your software but it looks that it also measures the Windows disk cache
which is nonsense. Try another program like the ATTO disk benchmark:
<http://members.home.nl/rvandesanden/Downloads/ATTO.rar>

Your writing results are indeed very low. Your disk write cache is
probably disabled, by enabling it you can increase the writing
performance (but you loose all data in write cache if there is a power
outage). But be aware that the writing performance of RAID5 is always
quite limited.

Just do a retest with the ATTO disk benchmark, set the Transfer Size to
0.5 to 1024kb and the total Length to 32MB.

Benjamin

Benjamin Gawert wrote:
> * :
>
> > I've just built a new home server; operating system is Server 2003 R2
> > standard. Motherboard is an Intel DQ965GF with a Core2Duo 1.86 and 2GB
> > DDR2 Kingston RAM.
> >
> > I've been doing some testing before posting so hopefully you'll be able
> > to help me out.
> >
> > In the system I have 3x500GB Seagate SATA drives and 1x320GB Maxtor
> > SATA drive. Initially I created a RAID 5 with the 3x500's. I
> > partitioned that into a 30GB for the system and that left somewhere
> > over 900 for the data.
> >
> > Once it was up and running I updated the drivers and then tested the
> > drive performance. I used YAPT. Read performance on both partitions
> > was as I had hoped, it was around 130meg/sec. Write performance on the
> > data partition was 90meg/sec. However write performance on the system
> > partition was only 10meg/sec. I've checked the net and can't find
> > anything and I've confirmed write cache is on.
> >
> > The next thing I did was to brake the RAID and have the drives as stand
> > alone. I installed Server on the 320GB maxtor and then tested
> > performance. The 3 Seagates all returned consistent 72meg/sec read and
> > write. The maxtor returned 62meg/sec read and write.
> >
> > I've now recreated the RAID 5 and installed Server again but all as 1
> > big partition. Read performance is back up to 130's,eg/sec but write
> > is still 10meg/sec.
> >
> > It's gone 1AM in the morning now but I might try XP rather than server
> > tomorrow but I'm sure that will be a waste of time.
>
> Yousuf already told you that you have a software RAID, so don't expect
> miracles. But usually a software RAID is quite fast on todays processors
> if the load isn't too big (and that shouldn't be the case on a home server).
>
> As to your problem, I think the results you get are bogus. 130MB/s with
> three 7200rpm SATA drives in RAID5 is a bit too optimistic. I don't know
> your software but it looks that it also measures the Windows disk cache
> which is nonsense. Try another program like the ATTO disk benchmark:
> <http://members.home.nl/rvandesanden/Downloads/ATTO.rar>
>
> Your writing results are indeed very low. Your disk write cache is
> probably disabled, by enabling it you can increase the writing
> performance (but you loose all data in write cache if there is a power
> outage). But be aware that the writing performance of RAID5 is always
> quite limited.
>
> Just do a retest with the ATTO disk benchmark, set the Transfer Size to
> 0.5 to 1024kb and the total Length to 32MB.
>
> Benjamin

Cheers Benjamin

Panic over. I did a test with IOmeter

Here are my result files:

RAID5
http://www.islandpcservices.co.uk/dan/misc/raid5.htm

Standalone Maxtor
http://www.islandpcservices.co.uk/dan/misc/maxtor.htm

Those were with a 1gig file and 2 mins run time. The speeds with bigger
tst files is consistent.

The RAID has some 300gig of data on it now so the test file would have
been created near the middle of the disc. The Maxtor only has 10gig on
it. I'd imagine if both drives were empty the speeds would be slightly
faster.

wrote:
>
> On Dec 21, 4:04 pm, Yousuf Khan <bbb...@yahoo.com> wrote:
>> mrdanielmor...@gmail.com wrote:
>>> Hi
>>> Apologies if this is posted in the wrong group, if so could you point
>>> me in the right direction.
>>> I've just built a new home server; operating system is Server 2003 R2
>>> standard. Motherboard is an Intel DQ965GF with a Core2Duo 1.86 and 2GB
>>> DDR2 Kingston RAM.It's typical for RAID5 write performance. Also remember that this is a
>> software RAID5 rather than a hardware RAID5, which makes it even worse.
>>
>> Yousuf Khan
>
> Software RAID? are you sure
>
Actually a firmware RAID, but essentially all that means is that you
have a slightly better chance of boot if you have some very unlikely
failure modes. It's not hardware, where you hand the data to the
controller and smile, the firmware breaks the data into chunks,
calculates parity, and issues a separate write to each drive (and if you
have PATA drives on the same cable that really sucks).

Having the write size larger than the stripe size helps, particularly if
the write is some multiple of what fits on a single trips (N-1 data
chunks + one parity).

RAID-5 is expensive for write.

--
Bill Davidsen
He was a full-time professional cat, not some moonlighting
ferret or weasel. He knew about these things.