|
|
|
date: Mon, 10 Dec 2007 22:15:11 -0000,
group: microsoft.public.exchange.design
back
Re: Exchange 2007 storage advice
On Mon, 10 Dec 2007 22:15:11 -0000, "Matt76-uk"
wrote:
>Hi,
>
>I posted this in the exchange.admin group a few days ago but did not really
>get any feedback, I post here in the hope that someone may like to discuss
>their experiences.
>
>We are looking at a 1500 user migration to exchange 2007. I am interested in
>peoples ideas so far as DB/LOG storage for CCR nodes would be.
>
>The options i have considered:
>
>Option1 (on each CCR node)
>HP Smart Array P800 controller (16 port/3G SAS)
>1 x HP MSA 60 (DBs + Hot Spares) (RAID10)
>1 x HP MSA 60 (Logs + Hot Spares) (RAID10)
>
>Option2 (on each CCR node)
>2 x HP SC fibre HBA
>1 x HP MSA1000 (DBs + Logs + HotSpare) (dual controller / dual port)
>
>I realise that there are other subsystems to consider, but assuming all
>things being equal, what do you think would give the best outright
>performance so far as the storage is concerned?
>
>I look forward to your feedback.
>
>Thanks
>Matthew
I saw Oliver gave you something, was that not it?
Anyhow... Doing CCR on an MSA is plain daft and a total waste of time
and money. There's no point creating a cluster on a shed of a box like
that. Don't waste your time. Make a standalone box and get down to the
pub early on a Friday for a change. Really, trust me.
RAID10 is just pointless for you, certainly on the MSA. Nah, it's just
a waste of time.
date: Mon, 10 Dec 2007 20:13:05 -0500
author: Mark Arnold [MVP]
Re: Exchange 2007 storage advice
Hi Mark,
Yes Oliver answered the original question. Unfortunately the ESRP advice for
a similar deployment is not available.
Whats wrong with the MSA1000? It is meant to be a no frills solution, easy
to configure, easy to expand. When I have deployed it (DC/multi shelf), it
seems to perform quite well, low latency etc. Granted this has been for
SCC/mailboxes server deployments with <1000 users, hence the question
regarding CCR deployments.
Do you think that DAS is the preferred option for CCR deployments?
Why is the RAID10 option a waste of time? I thought the following was true
with this technology. You gain more IOPS in less disks. Sustain disk
failures without a noticible hit. Can take more disk failures than RAID 5.
All i have ever read recommends RAID10 for these reasons, i am interested in
why you would not recommend this.
I look forward to your feedback.
Regards,
Matthew
"Mark Arnold [MVP]" wrote in message
news:hqorl3tqk1ec7mdoeoq0lhglo24v1caioh@4ax.com...
> On Mon, 10 Dec 2007 22:15:11 -0000, "Matt76-uk"
> wrote:
>
>>Hi,
>>
>>I posted this in the exchange.admin group a few days ago but did not
>>really
>>get any feedback, I post here in the hope that someone may like to discuss
>>their experiences.
>>
>>We are looking at a 1500 user migration to exchange 2007. I am interested
>>in
>>peoples ideas so far as DB/LOG storage for CCR nodes would be.
>>
>>The options i have considered:
>>
>>Option1 (on each CCR node)
>>HP Smart Array P800 controller (16 port/3G SAS)
>>1 x HP MSA 60 (DBs + Hot Spares) (RAID10)
>>1 x HP MSA 60 (Logs + Hot Spares) (RAID10)
>>
>>Option2 (on each CCR node)
>>2 x HP SC fibre HBA
>>1 x HP MSA1000 (DBs + Logs + HotSpare) (dual controller / dual port)
>>
>>I realise that there are other subsystems to consider, but assuming all
>>things being equal, what do you think would give the best outright
>>performance so far as the storage is concerned?
>>
>>I look forward to your feedback.
>>
>>Thanks
>>Matthew
>
> I saw Oliver gave you something, was that not it?
>
> Anyhow... Doing CCR on an MSA is plain daft and a total waste of time
> and money. There's no point creating a cluster on a shed of a box like
> that. Don't waste your time. Make a standalone box and get down to the
> pub early on a Friday for a change. Really, trust me.
>
> RAID10 is just pointless for you, certainly on the MSA. Nah, it's just
> a waste of time.
date: Tue, 11 Dec 2007 08:24:38 -0000
author: Matthew Millers
Re: Exchange 2007 storage advice
Hey, I have have one those. It's ancient, and only half populated with 72s,
has only one controller and an 8 port switch, but what heck; it does a nice
job of heating the home office. It's basically an FC attached RAID
controller & enclosure of the plain vanilla approach.
Seriously; before you ask a sizing question, you need to know someting
about the load you will be placing on the potential candidates. Go get the
latest version of the MS Exchange Sizing calculator, and use that to work up
your estimated IO requirement. Post that, and then it will be possible to
rationally consider potential solutions.
One thought, and I think this is to Arnold's point; with no requirement for
shared storage in an Exchange 2007 CCR cluster, why would you make the
investment in the FC infrastucture? Can you name some value add that the
MSA is giving you over DAS?
When you look at the HP EVA line, or Netapp, or even EMC, the virualization
layer presents a set of features that may add some value. On top of that,
the vendor may have specific replication or backup tools that leverage the
hardware providing some value add. I think
http://technet.microsoft.com/en-us/library/bb738146.aspx in discussing RAID
DP describes one of the advanages of the virtualization layer on Netapp;
there's no write penalty. You get RAID 6 like space with RAID 0 like
performance. A byproduct of never overwriting is that snapshots become very
inexpensive in terms of performance impact. You don't need to copy on
write to take a snapshot; the data is already there because you never
overwrote it. That's why you can have 250 snaps per volume in place with
negligable performance impact.
If it were me, I'd be looking for a feature sets like this. In addition,
I'd look for iSCSI support instead of FC. Then again, I work for Netapp so
I'm probably biased. Do take the time to look at all the vendors' options
with an eye to what value add they provide and how that impacts your
business objectives. If there is none, then you'd be better off going DAS.
"Mark Arnold [MVP]" wrote in message
news:hqorl3tqk1ec7mdoeoq0lhglo24v1caioh@4ax.com...
> On Mon, 10 Dec 2007 22:15:11 -0000, "Matt76-uk"
> wrote:
>
>>Hi,
>>
>>I posted this in the exchange.admin group a few days ago but did not
>>really
>>get any feedback, I post here in the hope that someone may like to discuss
>>their experiences.
>>
>>We are looking at a 1500 user migration to exchange 2007. I am interested
>>in
>>peoples ideas so far as DB/LOG storage for CCR nodes would be.
>>
>>The options i have considered:
>>
>>Option1 (on each CCR node)
>>HP Smart Array P800 controller (16 port/3G SAS)
>>1 x HP MSA 60 (DBs + Hot Spares) (RAID10)
>>1 x HP MSA 60 (Logs + Hot Spares) (RAID10)
>>
>>Option2 (on each CCR node)
>>2 x HP SC fibre HBA
>>1 x HP MSA1000 (DBs + Logs + HotSpare) (dual controller / dual port)
>>
>>I realise that there are other subsystems to consider, but assuming all
>>things being equal, what do you think would give the best outright
>>performance so far as the storage is concerned?
>>
>>I look forward to your feedback.
>>
>>Thanks
>>Matthew
>
> I saw Oliver gave you something, was that not it?
>
> Anyhow... Doing CCR on an MSA is plain daft and a total waste of time
> and money. There's no point creating a cluster on a shed of a box like
> that. Don't waste your time. Make a standalone box and get down to the
> pub early on a Friday for a change. Really, trust me.
>
> RAID10 is just pointless for you, certainly on the MSA. Nah, it's just
> a waste of time.
date: Tue, 11 Dec 2007 16:50:16 -0800
author: John Fullbright fjohn@donotspamnetappdotcom
Re: Exchange 2007 storage advice
Hi John,
Appreciate your assistance, I worked out the following with the calculator.
This goes back to the maybe not so greatly worded original question, for CCR
with the IOPs below, do you think that the preferred option is DAS or SAN
(assuming no existing SAN environment).
I have also detailed the config using a MSA70 (max 15k 146gb 2.5 SAS disk)
on a P800 controller (or 2). We are actually looking at using more netapp
equipment, I would be more than open for any netapp options you may have,
although it may be priced out as an option.
I have really over specced the IOPS/user and configured with online mode for
all users.
Number of Mailboxes / Server 2300 ( i have chosen
this amount to include 15% growth)
Send/Receive Capability / Mailbox / Day 20 sent/80 received
(this in the high usage user pattern)
Average Message Size (KB) 50
Tier-1 User Mailbox Size Limit (MB) 250 (this will
accommodate their online storage requirements)
Tier-1 User IOPS / mailbox 0.85 (high usage
user pattern)
Tier-1 Database Read:Write Ratio 75%
Outlook Mode (Majority of Clients) Online Mode
Number of Mailboxes / Database 575
Number of Data Copies 2
(CCR instances)
Total Database Disk Space Required / Replica 871 GB
Total Log Disk Space Required / Replica 218 GB
Total Database LUN Disk Space Required / Replica 1144 GB (this is
the total of all DB LUNs which are 300GB each)
Required Database IOPS / Replica 2346
(IOPS)
Required Log IOPS / Replica 484
(IOPS)
"Database Layout"
DB1 300GB 575mailbox/DB (600IOPS)
DB2 300GB 575mailbox/DB (600IOPS)
DB3 300GB 575mailbox/DB (600IOPS)
DB4 300GB 575mailbox/DB (600IOPS)
(2400 IOPS total)
SG1 90GB
(125IOPS)
SG2 90GB
(125IOPS)
SG3 90GB
(125IOPS)
SG4 90GB
(125IOPS) (500 IOPS total)
Restore LUN (SG/DB) 340 GB
"Estimated IOPs and required spindles"
Databases (per DB)
10k RPM disks - 78 IOPS/disk (2:1 ratio ; includes write hit for RAID10)
8 disks @ 78 IOPS = 624 IOPS (per DB LUN)
15k RPM disks - 108 IOPS/disk (2:1 ratio ; includes write hit for RAID10)
6 disks @ 108 IOPS = 648 IOPS (per DB LUN)
Storage Groups (per SG)
10k RPM disks - 78 IOPS/disk (2:1 ratio ; includes write hit for RAID10)
2 disk @ 78 IOPS = 156 IOPS (per log LUN)
15k RPM disks - 108 IOPS/disk (2:1 ratio ; includes write hit for RAID10)
2 disk @ 108 IOPS = 216 IOPS (per log LUN)
"Disk Layout / Shelf Configuration (based on required spindle count)"
Following are x2 requirements, one for each CCR node.
I was either thinking of having two P800 controllers or a single P800 and
running each shelf on seperate channels on the P800. From this I could split
the RAID10 sets across the shelves (which the controller support).
"Using HP 146GB 3G 10K SAS 2.5" SFF SP HDD 431958-B21 in MSA70 (2.5 in
support)"
MSA70 Unit 1 (Controller 1 (or controller 1/channel 1))
16 disks for DBs/RSG (DB1 / DB2) (RAID10-2LUNS) (16
x 146 / 2 = 1168GB usable space ; 550GB per LUN)
8 disks for logs/RSG (SG1 / SG2) (RAID10-2LUNS)
(8 x 146 / 2 = 584GB usable space ; 270GB per LUN)
1 disks for hotspare
MSA70 Unit 2 (Controller 2 (or controller 1/channel 2))
16 disks for DBs/RSG (DB1 / DB2) (RAID10-2LUNS)
(16 x 146 / 2 = 1168GB usable space ; 550GB per LUN)
8 disks for logs/RSG (SG1 / SG2) (RAID10-2LUNS)
(8 x 146 / 2 = 584GB usable space ; 270GB per LUN)
1 disks for hotspare
Thanks
Matthew
"John Fullbright" <fjohn@donotspamnetappdotcom> wrote in message
news:OehO3jFPIHA.5264@TK2MSFTNGP02.phx.gbl...
> Hey, I have have one those. It's ancient, and only half populated with
> 72s, has only one controller and an 8 port switch, but what heck; it does
> a nice job of heating the home office. It's basically an FC attached RAID
> controller & enclosure of the plain vanilla approach.
>
> Seriously; before you ask a sizing question, you need to know someting
> about the load you will be placing on the potential candidates. Go get the
> latest version of the MS Exchange Sizing calculator, and use that to work
> up your estimated IO requirement. Post that, and then it will be possible
> to rationally consider potential solutions.
>
> One thought, and I think this is to Arnold's point; with no requirement
> for shared storage in an Exchange 2007 CCR cluster, why would you make the
> investment in the FC infrastucture? Can you name some value add that the
> MSA is giving you over DAS?
>
> When you look at the HP EVA line, or Netapp, or even EMC, the
> virualization layer presents a set of features that may add some value.
> On top of that, the vendor may have specific replication or backup tools
> that leverage the hardware providing some value add. I think
> http://technet.microsoft.com/en-us/library/bb738146.aspx in discussing
> RAID DP describes one of the advanages of the virtualization layer on
> Netapp; there's no write penalty. You get RAID 6 like space with RAID 0
> like performance. A byproduct of never overwriting is that snapshots
> become very inexpensive in terms of performance impact. You don't need
> to copy on write to take a snapshot; the data is already there because you
> never overwrote it. That's why you can have 250 snaps per volume in place
> with negligable performance impact.
>
> If it were me, I'd be looking for a feature sets like this. In addition,
> I'd look for iSCSI support instead of FC. Then again, I work for Netapp
> so I'm probably biased. Do take the time to look at all the vendors'
> options with an eye to what value add they provide and how that impacts
> your business objectives. If there is none, then you'd be better off
> going DAS.
>
>
>
>
> "Mark Arnold [MVP]" wrote in message
> news:hqorl3tqk1ec7mdoeoq0lhglo24v1caioh@4ax.com...
>> On Mon, 10 Dec 2007 22:15:11 -0000, "Matt76-uk"
>> wrote:
>>
>>>Hi,
>>>
>>>I posted this in the exchange.admin group a few days ago but did not
>>>really
>>>get any feedback, I post here in the hope that someone may like to
>>>discuss
>>>their experiences.
>>>
>>>We are looking at a 1500 user migration to exchange 2007. I am interested
>>>in
>>>peoples ideas so far as DB/LOG storage for CCR nodes would be.
>>>
>>>The options i have considered:
>>>
>>>Option1 (on each CCR node)
>>>HP Smart Array P800 controller (16 port/3G SAS)
>>>1 x HP MSA 60 (DBs + Hot Spares) (RAID10)
>>>1 x HP MSA 60 (Logs + Hot Spares) (RAID10)
>>>
>>>Option2 (on each CCR node)
>>>2 x HP SC fibre HBA
>>>1 x HP MSA1000 (DBs + Logs + HotSpare) (dual controller / dual port)
>>>
>>>I realise that there are other subsystems to consider, but assuming all
>>>things being equal, what do you think would give the best outright
>>>performance so far as the storage is concerned?
>>>
>>>I look forward to your feedback.
>>>
>>>Thanks
>>>Matthew
>>
>> I saw Oliver gave you something, was that not it?
>>
>> Anyhow... Doing CCR on an MSA is plain daft and a total waste of time
>> and money. There's no point creating a cluster on a shed of a box like
>> that. Don't waste your time. Make a standalone box and get down to the
>> pub early on a Friday for a change. Really, trust me.
>>
>> RAID10 is just pointless for you, certainly on the MSA. Nah, it's just
>> a waste of time.
>
>
date: Wed, 12 Dec 2007 10:10:23 -0000
author: Matthew Millers
|
|
