Blog Post

Exchange Team Blog
3 MIN READ

Exchange 2013 Calculator Updates

Ross Smith IV's avatar
Ross Smith IV
Icon for Microsoft rankMicrosoft
Jun 19, 2015

Today, we released an updated version of the Exchange 2013 Server Role Requirements Calculator.

In addition to numerous bug fixes, this version includes new functionality: CPU utilization table, ReplayLagManager support, MaximumPreferredActiveDatabases support, Restore-DatabaseAvailabilityGroup scenario support, and guidance on sizing recommendations. You can view what changes have been made or downloadthe update directly. For details on the new features, read on.

CPU Utilization Table

The Role Requirements tab includes a table that outlines the expected theoretical CPU utilization for various modes:

  • Normal Run Time (where the active copies are distributed according to ActivationPreference=1)
  • Single Server Failure (redistribution of active copies based on a single server failure event)
  • Double Server Failure (redistribution of active copies based on a double server failure event)
  • Site Failure (datacenter activation)
  • Worst Failure Mode (in some cases, this value will equal one of the previous scenarios, it could also be a scenario like Site Failure + 1 server failure; the worst failure mode is what is used to calculate memory and CPU requirements)

Here’s an example:

Calc1

In the above scenario, the worst failure mode is a site failure + 1 additional server failure (since this is a 4 database copy architecture).

ReplayLagManager Support

ReplayLagManager is a new feature in Exchange Server 2013 that automatically plays down the lagged database copy when availability is compromised. While it is disabled by default, we recommend it be enabled as part of the Preferred Architecture.

Prior to version 7.5, the calculator only supported ReplayLagManagerin the scripts created via the Distribution tab (the Role Requirements and Activation Scenarios tabs did not support it). As a result, the calculator did not factor the lagged database copy as a viable activation target for the worst failure mode. Naturally, this is an issue because sizing is based on the number of active copies and the more copies activated on a server, the greater the impact to CPU and memory requirements.

In a 4-copy 2+2 site resilient design, with the fourth copy being lagged, what this meant in terms of failure modes, is that the calculator sized the environment based on what it considered the worst case failure mode – Site Failure (2 HA copies lost, only a single HA copy remaining). Using the CPU table above as an example, calculator versions prior to 7.5 would base the design requirements on 18 active database copies (site failure) instead of 22 active database copies (3 copies lost, lagged copy played down and being utilized as the remaining active).

ReplayLagManageris only supported (from the calculator perspective) when the design leverages:

  • Multiple Databases / Volume
  • 3+ HA copies

MaximumPreferredActiveDatabases Support

Exchange 2010 introduced the MaximumActiveDatabasesparameter which defines the maximum number of databases that are allowed to be activated on a server by BCS. It is this value that is used in sizing a Mailbox server (and is defined the worst failure mode in the calculator).

Exchange 2013 introduced an additional parameter, MaximumPreferredActiveDatabases. This parameter specifies a preferred maximum number of databases that the Mailbox server should have. The value of MaximumPreferredActiveDatabasesis only honored during best copy and server selection (phases 1 through 4), database and server switchovers, and when rebalancing the DAG.

With version 7.5 or later, the calculator recommends setting MaximumPreferredActiveDatabases when there are four or more total database copies. Also, the Export DAG List form exposes the MaximumPreferredActiveDatabasessetting and createdag.ps1 sets the value for the parameter.

Restore-DatabaseAvailabilityGroup Scenario Support

In prior releases, the Distribution tab only supported the concept of Fail WAN, which allowed you to simulate the effects of a WAN failure and model the surviving datacenter’s reaction depending on the location of the Witness server. However, Fail WAN did not attempt to shrink the quorum, so if you attempted to fail an additional server you would end up in this condition:

calcFailWAN

With this version 7.5 and later, the calculator adds a new mode: Fail Site.  When Fail Site is used, the datacenter switchover steps are performed (and thus the quorum is shrunk, alternate witness is utilized, if required, etc.) thereby allowing you to fail additional servers.  This allows you to simulate the worst failure mode that is identified in the Role Requirements and Activation Scenarios tabs.

calcFailSite

Note: In order to recover from the Fail Site mode, you must click the Refresh Database Layout button.

Sizing Guidance Recommendations

As Jeff recently discussed in Ask The Perf Guy: How Big Is Too Big?, we are now providing explicit recommendations on the maximum number of processor cores and memory that should be deployed in each Exchange 2013 server. The calculator will now warn you if you attempt a design that exceeds these recommendations.

cpu

As always, we welcome your feedback.

Ross Smith IV
Principal Program Manager
Office 365 Customer Experience

Updated Jul 01, 2019
Version 2.0
  • There is no link to download the calculator on the Gallery
  • Hi Ross, I think there is a bug in mailbox space modelling formula. Instead of "=(calcTier1YearlyData+(calcTier1InitialMBXSize/1024))*T12/12", it should be =(calcTier1YearlyData)*T12/12+(calcTier1InitialMBXSize/1024). If initial size is 5GB, yearly generated

    data 1GB, I suppose after 5 years with 100% retention, mailbox size should be 10GB (5GB+5x1GB). With current formula result is 30 Gb. It does not make any sense to me. Thank you. Mirek

  • rodthekiller's avatar
    rodthekiller
    Copper Contributor

    Hello, where i can found the download file ? Thank you for your help