Knowing What Tweaks You Can Make to the Microsoft Exchange Spreadsheet Calculator
Plug-n-play - that's part of the idea behind the Microsoft Exchange Storage Calculator spreadsheet which provides organizations the general guidelines that they need when planning and configuring the storage that will support an instance of Microsoft Exchange. However one should not assume that this spreadsheet takes into account every possible variable regarding storage systems - it most certainly does not and says as much within the spreadsheet. So before making any tweaks to the spreadsheet, one must know something about the capabilities of the storage system that one will use to host Exchange and what changes one can confidently make to the spreadsheet without botching an Exchange implementation.
The "Storage Design" page within the Exchange Calculator spreadsheet is the one that organizations will specifically want to take a hard look at when they start to plan for and configure the storage system that will host Exchange. At the top of this page, Microsoft provides some guidelines as to how much overhead a disk drive failure within a RAID group will introduce during rebuild times (35% for RAID 1/0 and 100% for RAID 5 or 6).
The motivation behind Microsoft providing these percentages is to assist organizations in appropriately sizing their storage system should a disk drive failure occur. These percentages ensure the storage system maintains optimal performance for Exchange even in the event of a disk drive failure. So as an organization interprets these percentages and applies it to their environment, this is the amount of extra capacity they must add to their storage system to maintain production levels of IOPs and throughput for Exchange.
In the case of RAID 1/0, an organization needs an additional one third (1/3) more capacity (35%) to ensure forecasted production levels of I/O are met during the rebuild of a failed disk drive. In cases where RAID 5 or RAID 6 are used on the storage system to support Exchange, organizations need to plan to implement twice the disk drives if they want to offset the 100% overhead that rebuilds of disk drives in these RAID configurations incur.
While these overhead percentages associated with these different RAID configurations appear logical on the surface (and they are), organizations should verify that these percentages are applicable to the storage system that they are looking to implement behind Exchange. Since Microsoft makes a number of assumptions in its spreadsheet about RAID technology, this is one area that organizations can look to tweak without negatively impacting their Exchange implementation.
Bill Plein, a storage architect with 3PAR's Strategic Accounts Engineering Team, explained that he works very closely with 3PAR customers to help them understand that they can satisfy all of Microsoft's best practices without needing to follow this particular detail on Microsoft's storage calculator spreadsheet. While 3PAR supports and follows all of Microsoft's best practices, he helps customers understand some recommendations around storage design are important for availability reasons and others for performance.
In the case of the 3PAR InServ Storage Server, it uses a wide striping technology so it does not have the same performance problems regarding the RAID rebuild overhead that Microsoft's recommendations are trying to address. Instead 3PAR spreads out its rebuilds across every drive in the system, not just to a single replacement drive.
So if a 3PAR InServ Storage Server has hundreds of drives in it and it looses one drive, almost every drive in the system get a little bit warmer because all of the drives start rebuilding the parity or the mirror chunklets that were lost when that one drive went down. Then when that drive is repaired, all of the data is leaked back to the new drive in a background process.
Since the drives participate one by one and restore the data in the background, 3PAR's rebuild time is fast and the performance penalty is extremely low - less than 5% if measurable at all. So even though Microsoft says allow for a 35% overhead on RAID 1/0 systems and 100% on RAID 5 and 6 systems, 3PAR can deliver the same reliability and performance using its wide striping technology as these other approaches with less than a 5% overhead on its system.
Plein concedes that 3PAR customers conservatively plan for a percentage higher than 5%. He says, "Even if the customer goes with an ultra-conservative ratio of 10 - 20% of overhead for RAID 5, 3PAR is basically selling its customers less disk as they do not need all of the Mirosoft-recommended spindles. Here is a place where 3PAR does it better and they can back off on the Microsoft recommendation."
In most organizations, Microsoft Exchange is too critical of an application both from a business and technical perspective to make any tweaks to Microsoft's recommended configurations without solid evidence that they will work as designed. However this is one example where Microsoft is over-compensating in its recommendation because it has to design its specifications to the lowest common denominator which, in this case, is traditional RAID arrays. Using 3PAR's wide striping technology in lieu of traditional arrays, the customer can be pretty confident that when the solution is rolled out there will be ample head room on the system without needing to purchase a lot of extra disk to accomplish that objective.
The "Storage Design" page within the Exchange Calculator spreadsheet is the one that organizations will specifically want to take a hard look at when they start to plan for and configure the storage system that will host Exchange. At the top of this page, Microsoft provides some guidelines as to how much overhead a disk drive failure within a RAID group will introduce during rebuild times (35% for RAID 1/0 and 100% for RAID 5 or 6).
The motivation behind Microsoft providing these percentages is to assist organizations in appropriately sizing their storage system should a disk drive failure occur. These percentages ensure the storage system maintains optimal performance for Exchange even in the event of a disk drive failure. So as an organization interprets these percentages and applies it to their environment, this is the amount of extra capacity they must add to their storage system to maintain production levels of IOPs and throughput for Exchange.
In the case of RAID 1/0, an organization needs an additional one third (1/3) more capacity (35%) to ensure forecasted production levels of I/O are met during the rebuild of a failed disk drive. In cases where RAID 5 or RAID 6 are used on the storage system to support Exchange, organizations need to plan to implement twice the disk drives if they want to offset the 100% overhead that rebuilds of disk drives in these RAID configurations incur.
While these overhead percentages associated with these different RAID configurations appear logical on the surface (and they are), organizations should verify that these percentages are applicable to the storage system that they are looking to implement behind Exchange. Since Microsoft makes a number of assumptions in its spreadsheet about RAID technology, this is one area that organizations can look to tweak without negatively impacting their Exchange implementation.
Bill Plein, a storage architect with 3PAR's Strategic Accounts Engineering Team, explained that he works very closely with 3PAR customers to help them understand that they can satisfy all of Microsoft's best practices without needing to follow this particular detail on Microsoft's storage calculator spreadsheet. While 3PAR supports and follows all of Microsoft's best practices, he helps customers understand some recommendations around storage design are important for availability reasons and others for performance.
In the case of the 3PAR InServ Storage Server, it uses a wide striping technology so it does not have the same performance problems regarding the RAID rebuild overhead that Microsoft's recommendations are trying to address. Instead 3PAR spreads out its rebuilds across every drive in the system, not just to a single replacement drive.
So if a 3PAR InServ Storage Server has hundreds of drives in it and it looses one drive, almost every drive in the system get a little bit warmer because all of the drives start rebuilding the parity or the mirror chunklets that were lost when that one drive went down. Then when that drive is repaired, all of the data is leaked back to the new drive in a background process.
Since the drives participate one by one and restore the data in the background, 3PAR's rebuild time is fast and the performance penalty is extremely low - less than 5% if measurable at all. So even though Microsoft says allow for a 35% overhead on RAID 1/0 systems and 100% on RAID 5 and 6 systems, 3PAR can deliver the same reliability and performance using its wide striping technology as these other approaches with less than a 5% overhead on its system.
Plein concedes that 3PAR customers conservatively plan for a percentage higher than 5%. He says, "Even if the customer goes with an ultra-conservative ratio of 10 - 20% of overhead for RAID 5, 3PAR is basically selling its customers less disk as they do not need all of the Mirosoft-recommended spindles. Here is a place where 3PAR does it better and they can back off on the Microsoft recommendation."
In most organizations, Microsoft Exchange is too critical of an application both from a business and technical perspective to make any tweaks to Microsoft's recommended configurations without solid evidence that they will work as designed. However this is one example where Microsoft is over-compensating in its recommendation because it has to design its specifications to the lowest common denominator which, in this case, is traditional RAID arrays. Using 3PAR's wide striping technology in lieu of traditional arrays, the customer can be pretty confident that when the solution is rolled out there will be ample head room on the system without needing to purchase a lot of extra disk to accomplish that objective.
Newsletter Signup
Leave a comment