Sanity check my disk layouts... (Full Version)

All Forums >> [Microsoft Exchange 2010] >> Installation



Message


mkent -> Sanity check my disk layouts... (9.Nov.2012 4:08:13 AM)

Hi All,

1st post here: please be kind :)

I'm about to add an Exchange 2010 server (hosting all roles) into our existing 2003 site. Then remove the 2003 server after a period. In the future I would probably like to add another 2010 server at our other site and use DAGs.

I have a HP DL370G7 server dual CPU, 24Gb Ram with 7 x 300Gb 15k SAS drives. My plan is to create a single array of 6 x 300 Gb drives RAID 1+0 with a hot spare.

I'm then thinking that I should create logical volumes on the array as follows:
1 - Windows Exchange Binaries - 200Gb
2 - Pagefile - 50Gb
3 - Database - 600Gb
4 - Logfiles - 50Gb

We have about 100 users (not using cached mode) and their mailbox sizes varies from a few 100Mbs to 20Gb in some cases. Currently our priv / pub is about 400Gb, but we're about to archive some of this off via mimecast.

Does this sound sensible?

Thanks,
Matthew




zbnet -> RE: Sanity check my disk layouts... (16.Nov.2012 10:26:03 AM)

These questions are dealt with on this technet page: http://technet.microsoft.com/en-us/library/ff622323.aspx

Pagefile should be RAM+10MB (for servers with >8GB of memory).

Historically, on Exchange servers, you separate the DB and the LOGs for the DB onto separate LUNS. I know Exchange 2010 has made huge strides forward in reducing the IOPS requirements of the storage used for hosting mail databases, but still, to break the rule by putting the DBs on the same LUN as the LOGS you need multiple copies of the data (ie a DAG), and you don't have that yet.

I know you'll argue and say that the performance of a multiple spindle RAID 1+0 array is good enough, but you're potentially compromising your recovery of Exchange if the total volume fails (as it will if you suffer from a simultaneous - or overlapping - two-disk failure). Separate your DB and its LOGS onto separate LUNS (ie separate RAIDed disk arrays) and you'd then need 4 simultaneous (or overlapping) disk failures to toast your server's data.

Why aren't your users using cached mode Outlook? This is the recommendation for Exchange 2010, I don't know why you'd not want to use it.

If your current server disk capacity isn't big enough to store all the mail for your users then you need bigger/more disks. archiving off some of the old mail to mimecast to me is defeating the purpose of having Exchange. Exchange 2010 has great support for big mailboxes, and also includes within the product you've bought the concept of archive mailboxes. Keep the mail within Exchange, much easier to support, and a less complex overall solution.

Finally, if you're planning on adding additional servers to make a DAG, don't forget that the database DB paths need to be identical across the DAG servers, so the storage decisions you make for this first server will dictate the setup of the other DAG servers (at least as far as DBs are concerned). If you only add a second server to the DAG, you still need both servers to run RAID for the DBs, JBOD is only a good idea for DBs when you have a minimum of 3 DB copies/DAG nodes.




troy12n -> RE: Sanity check my disk layouts... (26.Nov.2012 3:46:25 PM)

Some things:

1. When doing raid1 on HP smartarrays, there is no "hot spare". That functionality exists only on raid5 or raid6. At least on HP hardware, this is how they implemented it.

2. Why are you using 15k disks? By all accounts I have heard, that's an unnecessary expense on 2010. By using 10k SAS disks, you have the ability to use 600 or 800 gig disks, which HP dont make in 15k SAS enterprise disks (I dont think anyone does)

3. use cached mode unless you have some sort of situation where you cant (VDI or a terminal server environment)

4. It's always been a best practice to put logs and databases on different disks. It mainly used to be because of the very different data patterns of log disks versus DB's. Log disks (lun), gets written too all day, a very smooth, constant write pattern. Whereas, on a DB disk (lun), it is getting random reads and writes. To mix those two it was said to lead to queuing and paging while 1 operation waited for another.

5. 20 gig mailbox? There is no reason for that. You need to sack up and get control of your users. That is just sheer laziness. Biggest mailbox size limit in my org of 5000+ employees is 1 gig. 95% of my users have 500mb size limits.

6. For 100 users, I would add more ram to the server. 48 megs minimum. Part of how Exchange 2010 is able to get more use of lower speed disks is by caching, double or more your ram.




Page: [1]