Guidance Change: Calculating the Megacycles for Different Processor Configurations Formula
Published Sep 14 2010 11:37 AM 14K Views

As part of your Exchange 2010 planning process, one of the areas you must focus on is the Mailbox server design.  One component in that design process will be determining your CPU requirements.  To that end, we have published guidance that identifies the megacycles / mailbox for each message profile that exists, for both active and passive mailboxes.  This data was collected using a baseline system (2x4 core Intel Xeon x5470 3.33-GHz processors that yields 3333 megacycles per processor core, with a SPECint2006 rate value per core of 18.75).  Most of you, however, will not be deploying Exchange 2010 on the same hardware that we used to collect this data; therefore, you must perform a megacycle adjustment to determine the available megacycles your server design can support.

Recently we made a change with respect to how we normalize the available megacycles of server platforms with respect to our baseline system.  Initially, we adjusted megacycles per core with respect to the total megacycles of the system and derived this formula:

((New platform per core value) * (Hertz per core of new platform)) / (Baseline per core value) = Adjusted Megacycles per Core

However, that is not the correct way to look at normalizing the server platforms against our baseline. What we are actually interested in is how this impacts the active mailbox user count.  Therefore the new formula is:

((New platform per core value) * (Hertz per core of baseline platform)) / (Baseline per core value) = Adjusted Megacycles per Core

To understand this change, let’s look at an example solution that utilizes an 8 core x5550 2.66GHz (2660 megacycles/core or 21280 megacycles) processor architecture that has a SpecInt2006 rate value of 30.12 per core.

  • If we divide 30.12 by 18.75 = 1.6 or a 60% increase when compared to the baseline platform.
  • If we take 1.6 and apply that to 2660 megacycles we get 4256 adjusted megacycles per core or 34048 adjusted megacycles per server.  This is 60% increase when compared to 21280 megacycles.
  • If we take 1.6 and apply that to 3333 megacycles we get 5333 adjusted megacycles per core or 42664 adjusted megacycles per server.  This is nearly a 100% increase when compared to 21280 megacycles.

However, if you look at this from the mailbox perspective, you get the following (assume 200 message profile, so 4 megacycles / active mailbox):

  • Baseline system has 3333 megacycles/core x 8 cores = 26664 total megacycles.
  • 26664 total megacycles / 4 megacycles/active mailbox = 6660 active mailbox users.
  • New system using 34048 adjusted megacycles / 4 megacycles/active mailbox = 8512 active mailbox users, which is a 30% increase over the baseline.
  • New system using 42664 adjusted megacycles / 4 megacycles/active mailbox = 10666 active mailbox users, which is a 60% increase over the baseline.

If we refer back to our SpecInt2006 rate value, the difference between the baseline system and this example system is 60%.  Therefore, the formula that needs to be used to determine the adjusted megacycle per core value is:

((New platform per core value) * (Hertz per core of baseline platform)) / (Baseline per core value) = Adjusted Megacycles per Core

or

((New platform per core value) * 3333) / 18.75 = Adjusted Megacycles per Core

This guidance change is effectively immediately and is documented within the TechNet article, Mailbox Server Processor Capacity Planning.  The documentation within the Exchange 2010 Mailbox Role Requirements Calculator will also be updated to reflect this. 

For those of you that have already sized your systems using the old, invalid formula, don’t panic, your systems simply will have some extra megacycle capacity. The extra amount will depend on the difference between the MHz value of your processor and the baseline value of 3333 MHz.

Ross Smith IV
8 Comments
Not applicable
I believe this needs further explanation. For example, where did you get the 18.75 number from?
Not applicable
Jason,

The baseline system has a SPECint2006 rate value of 150, therefore, each core has a value of 18.75; this is the "Baseline per core value" component in the formula.   This is explained in TechNet article that is referenced.

Ross
Not applicable
So I googled the word "megacycle" and this article is the 1st non-bicycling/motorcycling related result :)
Not applicable
So, based upon the referenced system, you are stating that the server can support 10,000+ mailboxes?
Not applicable
Exchange Noob - yes, from purely a CPU perspective, this server platform could support active 10,000 200 message profile mailboxes while at a 100% CPU utilization.  The recommendation would be to scale back the solution to no more than 80% CPU utilization (otherwise your server and users are in for a world of hurt), so this would be more like 8500 200 message profile mailboxes.

However, you still need to think about storage, memory, db design, multi-role, and network impacts, all of which, may decrease the active mailbox count. Not to mention, this would be the number of active mailboxes during the worst failure mode, not during normal run time operations.

Ross
Not applicable
When is an updated Mailbox Server Role Requirements Calculator going to be released now that Exchange 2010 SP1 is out? waiting.....
Not applicable
"don't panic, your systems simply will have some extra megacycle capacity. The extra amount will depend on the difference between the MHz value of your processor and the baseline value of 3333 MHz"

.. or will have some megacycle shortage when running CPUs clocked over 3,33 GHz
Not applicable
This is good article but still i have confusion here; if you can clarify it would be great...
In above formula ((New platform per core value) * 3333) / 18.75 = Adjusted Megacycles per Core), here SPECint2006 rate value per core of 18.75 is for the baseline system HP DL380 G5 x5470 3.33GHz, 8 cores (3,333 MHz) as per Microsoft TechNet article Mailbox Server Processor Capacity Planning.... Is this 18.75 value standard & we can take this 18.75 value as standard against all other new proceesor platform which actually is being used for the deployment..or it's differ case by case...there is also baseline value which results in SPECint2006...what is that..
Version history
Last update:
‎Jul 01 2019 03:54 PM
Updated by: