Exchange + RAID Sanity Check

Forums › IT Pro and developers › Exchange + RAID Sanity Check

Dynamic

3869 posts

Uber Geek

ID Verified

Trusted

Lifetime subscriber

#170996 1-Apr-2015 15:17

Just ordered a server for a client with 8 SFF disk bays. It has a pair of 300gb 15k disks fitted with the bundle including Server 2012R2, and I have ordered 4 more 300gb 10k disks (the 15k disks were notably more and exceeded budget).

Client is a professional services firm with 3 staff. They insisted they wanted to keep Exchange in-house and were prepared to pay for it.

Mirror the 15k disks for the OS and RAID10 the 10k disks for Exchange and File/Print? Or 6 disk RAID10 for the whole lot?

Current use:
60Gb Exchange database has 7 years data in it with all email to be preserved for 10 years as they have liability that far back for their services.
220Gb file storage also accumulated over the last 7 years, so the growth rate is not huge. They don't deal with multimedia files - it's all Office docs and PDFs.

I'm leaning toward the 6 disk RAID10. 2 disks will deliver the data faster than the rest, but that is a not a problem. Mixing disk speeds in an array is supported. 100Gb OS partition, 800Gb for data. If we went with the 4 disk RAID10 for data, 600Gb of space might be a bit tight in 3-4 years time. Some files could be shuffled to the OS disks which will have heaps pf space.

I was leaning toward a pair of 800Gb Intel SSDs (the good 3500 series models) which were not much more that the 4x 10k drives, but chickened out with potential warranty concerns from the server manufacturer.

Any thoughts?

Cheers.

“Don't believe anything you read on the net. Except this. Well, including this, I suppose.” Douglas Adams

Referral links to services I use, really like, and may be rewarded if you sign up:
PocketSmith for budgeting and personal finance management. A great Kiwi company.

wasabi2k

2096 posts

Uber Geek

#1275477 1-Apr-2015 15:55

a 60GB Exchange Store will not tax a RAID6 setup with SAS disks.

Just go RAID6 for additional capacity.

As always assuming you will have decent backups.

Dynamic

3869 posts

Uber Geek

ID Verified

Trusted

Lifetime subscriber

#1275484 1-Apr-2015 16:06

wasabi2k that's a smart suggestion that I had not considered. Extra capacity and allows for the failure of any two drives. I'm sure the 2Gb RAID card will cope with that. :)

“Don't believe anything you read on the net. Except this. Well, including this, I suppose.” Douglas Adams

Referral links to services I use, really like, and may be rewarded if you sign up:
PocketSmith for budgeting and personal finance management. A great Kiwi company.

Mark

1653 posts

Uber Geek

#1275869 2-Apr-2015 09:03

https://technet.microsoft.com/en-us/library/ee832792(v=exchg.150).aspx

Don't put logs and DB on the same physical disks/RAID set, with what you've bought best you can do is probably :

OS = 2 * 300GB in RAID 1
Logs = 2 * 300GB in RAID 1
DB = 2 * 300GB in RAID 1

Dynamic

3869 posts

Uber Geek

ID Verified

Trusted

Lifetime subscriber

#1275877 2-Apr-2015 09:12

Thank you for the thoughts, Mark.

Performance should not be an issue with the multiple spindles. Certainly in a larger Exchange installation I would split the logs and database.

3 chunks of 300Gb each would be a PITA to manage as this server will be hosting 220Gb of shared files as well.

I'll likely go 150Gb for the OS, 250Gb Exchange partition (database only), 500Gb data and Exchange logs, and leave 300Gb unallocated for future use.

“Don't believe anything you read on the net. Except this. Well, including this, I suppose.” Douglas Adams

Referral links to services I use, really like, and may be rewarded if you sign up:
PocketSmith for budgeting and personal finance management. A great Kiwi company.

Mark

1653 posts

Uber Geek

#1275891 2-Apr-2015 09:41

OK .. but you would do well to point out to the customer that if they have a RAID failure (multiple disk, or controller goes nuts or <insert one of many reasons here>) they will be restoring everything from their backups and all data in between when the backup was run and the failure occurring will be gone.

Log + DB separation is not just because of performance (though it does help a lot), it's because if you lose the DB you can restore from backup and roll in the logs. If you lose the logs you don't really care as the data is in the DB and new logs can be started.
Logs and DB on the same spindles is fine if you have a HA Exchange setup, but with standalone it's a bit of a no-no and you'll end up with angry customer if anything bad happens (especially if they are liable for the data ... I wonder if they would then try making you liable for not providing a solution that meets their requirements ?)

Why would it be a pain to manage 3 RAID sets ? Once they are setup they are setup nothing to do after that really other than make sure they stay healthy.

I'd also not mix spindle speeds in a RAID set, even if it is "supported", it would just give a weird performance profile.

wasabi2k

2096 posts

Uber Geek

#1275936 2-Apr-2015 10:05

Let's be realistic team.

Yes best practice is to separate DB and logs for both performance and restore. We do that here at work, but we also have SAN storage across multiple sites with thousands of mailboxes.

This is a 3 person consultancy.

Performance is NOT going to be an issue, particularly with newer versions of Exchange.

Playing back logs is great, assuming they haven't turned on circular logging. Depending on the nature of the business and their level of IT investment - they probably don't have monitored daily backups to Tape, so circular logging is again probably the easiest option, with full daily/weekly backups and a RPO of the last night's backup.

Without circular logging - inattention to backups can mean filling up your drives with log files that haven't been rolled up as there have not been recent backups.

For a small business, extra disk space and a reduced future spend (due to lack of space) is far more valuable than following best practice, particularly when the benefits of doing so would not be meaningful to their bottom line.

Mark

1653 posts

Uber Geek

#1275960 2-Apr-2015 11:01

I've lost track of how many customers overrule "best practices" because of a few extra dollars up front spend, but then jump up and down when things go south.

If you are going to do a job, you do it right.

Still, not my customer so go for your life and pick and chose which best practices you agree with ... just remember those best practices are written by the people who developed the product, they are not just made up for fun :-)

Quic Broadband

Move to New Zealand's best fibre broadband service (affiliate link). Free setup code: R587125ERQ6VE. Note that to use Quic Broadband you must be comfortable with configuring your own router.

Dynamic

3869 posts

Uber Geek

ID Verified

Trusted

Lifetime subscriber

#1276042 2-Apr-2015 12:54

Mark, thank you for your input. I appreciate your considered opinion.

Best practise is best practise. It is the ideal setup. Not all sites allow for that ideal to be realised. This site is kept very well backed up.

Some business get away with telling clients 'it's my way or the highway'. Some don't. If it were a 20 client site, I would push for an ideal setup. With a 3 client site that cannot be justified.

At a Microsoft seminar at Sky City 5-ish years ago that I attended, they had a US-based executive speaker who encouraged Microsoft Partners to use their heads. If the customer situation does not justify a high performance fully redundant disk array, don't do yourself out of a sale by putting this forward as the only option. Perhaps put it forward as an ideal option for them to work toward next time and explain the benefits, but don't lose the sale by being pig-headed about it. (I cannot promise this is recited word for word.)

We have had clients with RAID controllers go south, taking the array with them, and got them going in a reasonable timeframe from backups. Fortunately this is rare and fortunately for me both times in the last 10 years this has been with a brand of server that I have specifically documented to the customer that we do not recommend (but they proceeded anyway and had us manage the servers). Next time both clients followed my recommendations.

“Don't believe anything you read on the net. Except this. Well, including this, I suppose.” Douglas Adams

Referral links to services I use, really like, and may be rewarded if you sign up:
PocketSmith for budgeting and personal finance management. A great Kiwi company.

Ragnor

8223 posts

Uber Geek

Trusted

#1278947 8-Apr-2015 12:16

Dynamic:

I was leaning toward a pair of 800Gb Intel SSDs (the good 3500 series models) which were not much more that the 4x 10k drives, but chickened out with potential warranty concerns from the server manufacturer.

Any thoughts?

What's the budget? Either way SSD's are cheap enough these days that the price/performance is almost a no brainer.

Say you went 4x 200GB SSD or 4x 400GB if need be in RAID10, this would give you way way way better read/write performance and good fault tolerance.

Performance matters imo.

Mark: https://technet.microsoft.com/en-us/library/ee832792(v=exchg.150).aspx

Don't put logs and DB on the same physical disks/RAID set, with what you've bought best you can do is probably :

OS = 2 * 300GB in RAID 1
Logs = 2 * 300GB in RAID 1
DB = 2 * 300GB in RAID 1

Matters for spinning rust platters in a standalone server but it's 2015 an array of SSD's is so fast you don't need to run two separate arrays, disk i/o won't be the bottleneck.

Also not in the OP's case but... in many cases Exchange would be provisioned as a VM's connected to a SAN/NAS/DAS where that advice would also not apply.

Ragnor

8223 posts

Uber Geek

Trusted

#1278960 8-Apr-2015 12:29

wasabi2k: a 60GB Exchange Store will not tax a RAID6 setup with SAS disks.

Just go RAID6 for additional capacity.

As always assuming you will have decent backups.

Disk is relatively cheap, go RAID10... write performance is always terrible on RAID6

Dynamic

3869 posts

Uber Geek

ID Verified

Trusted

Lifetime subscriber

#1278968 8-Apr-2015 12:38

The ideal to have Exchange DB and Exchange Logs on separate storage is still an ideal scenario for if things REALLY hit the fan. When we have had RAID go bad, it has been the card's fault and arrays have not been recoverable. Does this mean with the ideal scenario you need twin RAID cards and separate hot-swap drive bays? So working toward the ideal for SMB's is the practical approach and keep the server well backed up.

We went RAID6 and the performance has been more than acceptable. 2Gb FBWC P420 RAID card seems to have heaps of horsepower.

We went ML350 G9 as it was on a decent special until 31 March... sheesh they are huge now with the space for 6 drive cages (of 8 drives each) in the front.

“Don't believe anything you read on the net. Except this. Well, including this, I suppose.” Douglas Adams

Referral links to services I use, really like, and may be rewarded if you sign up:
PocketSmith for budgeting and personal finance management. A great Kiwi company.

Mark

1653 posts

Uber Geek

#1279405 9-Apr-2015 07:06

Ragnor:

Mark: https://technet.microsoft.com/en-us/library/ee832792(v=exchg.150).aspx

Don't put logs and DB on the same physical disks/RAID set, with what you've bought best you can do is probably :

OS = 2 * 300GB in RAID 1
Logs = 2 * 300GB in RAID 1
DB = 2 * 300GB in RAID 1

Matters for spinning rust platters in a standalone server but it's 2015 an array of SSD's is so fast you don't need to run two separate arrays, disk i/o won't be the bottleneck.

Also not in the OP's case but... in many cases Exchange would be provisioned as a VM's connected to a SAN/NAS/DAS where that advice would also not apply.

It's not just disk contention I was avoiding by suggesting OS, Log and DB separation.