• Brief History of Exchange Sizing
  • Sizing come from production deployments
    • Dogfood, MSIT, customer and field feedback
  • Final IOPS comes from internal deployments with isolated user proviles
  • Guidance is conservative due to deployment scenarios that are different from Microsoft
  • Best Practices always includes IOPS validation with Jetstress, some customers use Loadgen for better system validation
  • MBX Role Requirements Calculator is the BEST way to size Exchange 2010
• Impact of 2013 Architecture
  • Single MBX role provides:
    • De-facto partition unit
    • Unit of High Availability
    • Cache Efficiencies
    • Hardware Efficiencies
    • Simialr to 2010
      • Must factor in CPU, Memory, Storage, and Network
  • Mailbox Servers
    • Memory requirements have increased in Exchange 2013
      • Mailbox or ‘all in one’ must have at least 8GB
    • Minimum CPU requirements follow published OS guidelines
    • Disk space requirements on install drive increased dramatically
      • ~30GB
      • Don’t forget to factor in your page file!
        • Memory Size + 10MB
    • Storage Sizing
      • Storage is sized for capacity as well as IOPS
      • Transport has storage requirements, may be sufficient to place transport queue on OS volume, consider space requirements of max queue growth
      • Safety Net results in additional storage needs for transport queue
        • Will dynamically increase this based on mail flow profiles and message sizing
        • Can move this off to dedicated drives/elsewhere if needbe
        • Store IOPS requirements reduced compared to prior versions
        • High availability design in a key factor in both capacity and IOPS sizing
  • CAS Role
    • Memory sizing
      • Significant memory requirements
      • Memory scales based on connection counts
      • MS guidance will likely scale memory with allocated CPU cores
      • Final guidance for 2013 likely equal to or a bit above current CAS 2010 guidance
    • CPU Sizing
      • CPU core ratio for determining required cas cpu resources when rtm guidance is available
      • Exchange 2010 guideline was 3:4, architecture change in Exchange 2013 reduces this
      • High availability likely a major factor in the number of required CAS servers
      • Disable hyperthreading (SMT) for the same reasons discussed for the Mailbox role
        • Reason:  A lot of managed processes, memory is allocated based on the number of cores, including the number of HT cores, so on allocation, more memory is allocated than is realistically needed and negatively impacts the system
        • Significant impact to some Exchange service memory footprints
        • SMT provides gain in processor throughput, but overall the gain is not worth the ‘cost’ based on our lab measurements
        • Virtualized environment:
          • VM will see the number of CPUs that are exposed
  • Aim for balanced hardware
    • Lowest TCO usually involves getting the most out of your hardware
    • Consider a 2U server with 12 LFF disks as a well-balanced hardware playform for the mailbox role
      • 2 drives R1 for OS and transport
      • ~9 drives for mailbox databases, no raid
      • ~1 drive as a spare
  • Sizing without guidance and tools
    • Use the balanced hardware as a starting point
    • Be conservative
    • Size for high availability requirements (failure domains!), then migrate slowly while monitoring
    • Add more hardware as necessary
    • Or, optionally, wait for guidance!
• Gist
  • Mailbox
    • Increasing CPU slightly over 2010
    • Increasing RAM a good bit over 2010
    • Storage is a 50% reduction over 2010
  • CAS
• Tools
  • Mailbox Role Requirements calculator/
    • Released at/after RTM
  • Jetstress
    • Available at RTM + 3 Months
    • ESRP
      • For storage partners, same deal, +3 months
  • Loadgen
    • RTM + 3 months

The biggest deal brought from the sizing session was the 3 month number.  So much that I verified with the speakers that those numbers were accurate.

If you are not willing to guess at your hardware (or just spend a ton and make sure you don’t have to do so), you will basically have to wait until those tools are released.