Last week, I went through a very interesting discussion with one of my customers who is implementing one of the largest Exchange cloud in the region, they are currently in the planning phase.
the discussion occurred during our design session planning for the overall implementation, the discussion started with the question of Storage design and RAID level required.
the storage that were calculated to host the projected hosted mailboxes were about 26 TB of data for 2 copies (around 13 TB) per copy.
the question that we got, now we will need to design our RAID storage, how many disk is needed? we will need a lot of disks!!!!
I paused a little and said, Why you will need to do RAID, there is actually a better way that will save you money and even offer better and more services to you customer, I paused “as part of the excitement process “ and said “Use 3 copies of Exchange 2010”.
my customer has a reputable SAN from a very reputable vendor, they have shelves and money to buy the shelves, but yet each penny you add to the solution will affect the final offering to the customers which is correct and true.
let us take for example, suppose that you want to host 10,000 users with 512 MB quote, and suppose that your Exchange factors will maintain the same factors you will do for Enterprise use (Like Deleted Items retention, overheads…etc). if you used the Exchange calculator this will lead you to need around 10 TB per copy (DB + Logs) and 20 TB per environment, and total required 1200 IOPs. ( to reach the same calculation, use the Exchange Storage Calculator).
of course I am not talking about the penetration factor and over-subscription factors that you as ISP might use during the calculation, you might assume that you have 10% of concurrency and usage or you might go with 90% this will be totally based on your marketing and strategy teams decision.
so to host the 10,000 users , 10 TB of data on RAID 5 or even RAID 10, let us see how many disk is needed.
using http://www.wmarow.com/strcalc/goals.html you will be able to set the storage and required IOPs, assuming 15 K, 450 GB disks, you will find that to accommodate all of the databases on single big RAID 5 LUN you will need it to be generated from 36 Disks which provides the required storage and around 2471 IOPs (obviously there is a lot of waste in the IOPs)
same calculation can be done with SATA Disks and that will leave you with about 21 * 1 TB disks which leaves you with 619 IOPs (limited set of IOPs) with about 13 TB of storage.
now let us go back to my suggestion, my suggestion was not to use any RAID level protection, using SATA disks and 3 Exchange nodes hosting the Exchange 2010 Databases, let us investigate the options:
|Using 15 K, 450 GB Disks||Using 1 TB SATA Disks||Using 3 Nodes Copy (single Disk 2 TB per DB)|
|Usable Storage||10058 GiB||14901 GiB||1 TB|
|No. of Disks Per Copy||36||21||14|
|Total Disks for environment||72||42||42|
|User Quote||512 MB||512 MB||512 MB|
|Possible Increase in quote||None||250 MB||250 MB|
|No. of Users per 1 TB storage (max recommended DB size)||1562||1200||1400|
|Max. No of supported users||10,000||10,000||14,000 (this is based on the calculation that assumes 512 MB with 1.25 overheads)|
The Possible Increase in quote calculated by available space that still remains on the disk divided by no. of users hosted per LUN which is 1 TB size.
for SAS/FC disks the maximum users per DB is based on the recommendations of max of 1 TB DB size, size overheads are set to 1.25, so space and max. users is limited with available usable space.
for SATA disks, the max of users determined and limited by max of IOPs that could be generated from the RAID Group.
it is much clearer for you know that with the use of 3 copies, you might have the same amount no. of disks used with 1 TB RAID 5, but you will enjoy larger mailboxes that you can offer for your customers without adding any cost to your investment or have more users capacity without sacrificing performance, adding the capacity to remove backup need and much simplified storage management (with the option of SAN elimination).
Consider the following scenario:
you have Exchange 2010 RTM infrastructure up and running, you applied SP1 to the servers, after applying Exchange 2010 SP1 you lost access to the OWA Virtual Directory, when accessing the OWA virtual directory you might get:
Internet Explorer cannot display this page
Could not load file or assembly ‘Microsoft.Exchange.Diagnostics, Version=18.104.22.168, Culture=neutral, PublicKeyToken=31bf3856ad364e35′ or one of its dependencies. The system cannot find the file specified.
when opening the OWA virtual director application path you get “invalid application path”
there are several root causes of this issues, below are my findings for the cause of these errors:
you are using wrong redirection configuration to redirect http to https traffic, the correct way to configure Exchange 2010 to redirect http to https traffic could be found here
using other techniques will lead to OWA loading errors.
Wrong configuration in the OWA web.config application, to correctly load it rename the file to web.config.old and restart iis.
in-complete SP1 or RU installation, try to remove SP1/RU and install it again.
I found the above issue on servers that had IE9 installed, didn’t have the chance to test whether uninstalling IE9 fixed the issue or not, as un-installing IE9 was taken among other steps to fix the issue, but if you might want to take a look to IE9.
[Blog Post] Timeout error when Trying to connect to a disk drive over iSCSI using #Netapp #snapdrive
Consider the following scenario:
you have a Netapp SAN, you created a FlexVol and Created a SAN LUN to be accessed from a system over iSCSI, when you connect to the LUN from the Snapdrive from the iSCSI initiator you get the following error
Error code : A timeout of 120 secs elapsed while waiting for volume arrival notification from the operating system.
if you create the LUN from the System Manager or the CLI, the Drive MUST be formatted first before adding it using Snapdrive, otherwise you must create the LUN using the Snapdrive wizard.
to solve this issue:
create the LUN using the snapdrive wizard or following the following steps:
1- assign the LUN/s to the server that will access the LUN/s by assigning initiator group to the LUN
2- format the drive and don’t map any drive letter to it.
3- disconnect the LUNs from the host by removing the LUN mapping to the initiator group.
4- connect to the drive using the snapdrive software.
it worth mentioning that it the best practice is to create the LUNs from the snapdrive interface unless you really need to pre-create the LUNs.