EMC Clariion

My company is considering the purchase of an EMC Clariion unit for our
OnDemand system. Has anybody implemented one recently (in the last
year) and do you have success or failure stories?

Thanks,

John

what kind of host?

John M wrote:
My company is considering the purchase of an EMC Clariion unit for our
OnDemand system. Has anybody implemented one recently (in the last
year) and do you have success or failure stories?

Thanks,

John

Hey John,

We implemented a Clariion with ATA disks and were not impressed with
performance. According to our CE it is due to the ATA disks and the
translation from FC (we're atached using Fibre) to the ATA that is the
bottleneck. He stated that we would see much better performance if we would
be using FC disks due to not having to do this "translation".
Hope this helps.

"John M" wrote in message
om...
My company is considering the purchase of an EMC Clariion unit for our
OnDemand system. Has anybody implemented one recently (in the last
year) and do you have success or failure stories?

Thanks,

John

Erik Hendrix wrote:
Hey John,

We implemented a Clariion with ATA disks and were not impressed with
performance. According to our CE it is due to the ATA disks and the
translation from FC (we're atached using Fibre) to the ATA that is the
bottleneck. He stated that we would see much better performance if we would
be using FC disks due to not having to do this "translation".
Hope this helps.

Indeed ATA disks by themselvs are not that slow. But the interfaces,
RAID controllers if you will, are a bottleneck, at least for the 'made
in Taiwan' models.

And I wouldn't be surprised if the 'made in Taiwan' shelves are used by
the big-name vendors.

My experience with 32 ATA spindles and 2 controllers in RAID5 versus 14
FC disks with software RAID and JBOD is that FC is faster by a factor of
5. I guess this will not even out if RAID5 is dropped, and so on. And
this is application with random reads, random and sequential writes,
where reading turns quickly in a bottleneck for the ATA disks.


Thomas

(John M) wrote in
om:

My company is considering the purchase of an EMC Clariion unit for our
OnDemand system. Has anybody implemented one recently (in the last
year) and do you have success or failure stories?

We're running four CX600's, a CX400 and have a couple of CX700's on the
way. Fairly nice gear at a decent price with only minor complications. It's
far from being an enterprise box (HDS 9980V or Symmetrix...) but it's good
enough [TM] for most of what we do and performs very well.

We have a single shelf of ATA drives which work out just fine, the rest (~
45-50TB) is FC.


--
/Jesper Monsted

"Erik Hendrix" wrote:

We implemented a Clariion with ATA disks and were not impressed with
performance. According to our CE it is due to the ATA disks and the
translation from FC (we're atached using Fibre) to the ATA that is the
bottleneck. He stated that we would see much better performance if we would
be using FC disks due to not having to do this "translation".
Hope this helps.

Would you be so kind to put your experience in aprox. numbers?
Max. sustained throughput in sequential read?
Max. no. of I/O/sec?

I get about 4 MB/sec (!) while copying big files. This is miserable.
Should be in the range 30-50 MB/sec.

Bye
Rick Denoire

Hey Rick,

We are striping over 40 a total of 40 disks (RAID 5). 20 disks/SP in the
Clariion and then striped on the host over both SP's. Our IO size is 4MB
(and is nicely striped down so that each IO request will hit every disk).
We are getting about 40MB/s total on the Clariion with the ATA disks.
4MB/sec sounds to me like you might just be going to 1 disk.

"Rick Denoire" wrote in message
...
"Erik Hendrix" wrote:

We implemented a Clariion with ATA disks and were not impressed with
performance. According to our CE it is due to the ATA disks and the
translation from FC (we're atached using Fibre) to the ATA that is the
bottleneck. He stated that we would see much better performance if we
would
be using FC disks due to not having to do this "translation".
Hope this helps.

Would you be so kind to put your experience in aprox. numbers?
Max. sustained throughput in sequential read?
Max. no. of I/O/sec?

I get about 4 MB/sec (!) while copying big files. This is miserable.
Should be in the range 30-50 MB/sec.

Bye
Rick Denoire

"Erik Hendrix" wrote:

We are striping over 40 a total of 40 disks (RAID 5). 20 disks/SP in the
Clariion and then striped on the host over both SP's. Our IO size is 4MB
(and is nicely striped down so that each IO request will hit every disk).
We are getting about 40MB/s total on the Clariion with the ATA disks.
4MB/sec sounds to me like you might just be going to 1 disk.
´
Your case sounds to me like you are not getting the performance that
you deserve in view of a 40 disk array. Read the specs, it states that
one disk is able to transfer up to 60 MB/sec.

Are all 40 disks in one RAID group? You should get at least (30 x No.
of disks in the RAID group) MB/sec.

Do I understand correct that one I/O operation with 4MB in size will
transfer (4/No._of_disks_in_the_RAID_group) MB, that means 100KB/disk,
assuming a RAID group of 40 disks, right?

These ATA disks are challenging, really.

Bye
Rick Denoire

Hey Rick,

We have 8 raid 5 groups (4+1). The stripe is 128KB.
We have 4 raid groups / SP.
On each SP, we've taken these 4 raid groups to create a metalun. We do a
stripe there of 512KB.
On the host we offer up 2 devices (1 from each SP) and then stripe over
these 2 devices (2MB stripe).
Our application does IO requests of 4MB. This results in:

4MB - 2 MB/device (=2MB/SP=2MB/MetaLun) - 512KB/Raid Group(in MetaLun) -
128KB/Disk

Thus each disk will receive a 128KB request for each 4MB request from the
application.

We only get 40MB/s throughput total on the Clariion. We were told that the
bottleneck is the translation from FC to ATA and that nothing can be done
about it. We can not add more "controllers" for this or more disks or
anything else.

But this just does not sound normal to me.

Any advice on figuring out where the bottleneck might be and how to resolve
it would be great.


"Rick Denoire" wrote in message
...
"Erik Hendrix" wrote:

We are striping over 40 a total of 40 disks (RAID 5). 20 disks/SP in the
Clariion and then striped on the host over both SP's. Our IO size is 4MB
(and is nicely striped down so that each IO request will hit every disk).
We are getting about 40MB/s total on the Clariion with the ATA disks.
4MB/sec sounds to me like you might just be going to 1 disk.
´
Your case sounds to me like you are not getting the performance that
you deserve in view of a 40 disk array. Read the specs, it states that
one disk is able to transfer up to 60 MB/sec.

Are all 40 disks in one RAID group? You should get at least (30 x No.
of disks in the RAID group) MB/sec.

Do I understand correct that one I/O operation with 4MB in size will
transfer (4/No._of_disks_in_the_RAID_group) MB, that means 100KB/disk,
assuming a RAID group of 40 disks, right?

These ATA disks are challenging, really.

Bye
Rick Denoire

Jesper Monsted wrote:

(John M) wrote in
. com:

My company is considering the purchase of an EMC Clariion unit for our
OnDemand system. Has anybody implemented one recently (in the last
year) and do you have success or failure stories?

We're running four CX600's, a CX400 and have a couple of CX700's on the
way. Fairly nice gear at a decent price with only minor complications. It's
far from being an enterprise box (HDS 9980V or Symmetrix...) but it's good
enough [TM] for most of what we do and performs very well.

We have a single shelf of ATA drives which work out just fine, the rest (~
45-50TB) is FC.

Would you report performance data of your cx systems?

I just one to know if it my box which is broken or EMC itself.

Bye
Rick Denoire