Hello,
I recently deployed an Exchange 2013 SP1 CU5 server and I am in the process of migrating all of my Exchange 2010 users to this server.
There is no DAG, single server and its multi-role that hosts about 300 employee mailboxes that use about 6TB of space split between archive/live databases. Within our environment our average message size is large due to attachments (nature of business) thus the large used space.
(there are about +200 shared mailboxes for misc. stuff but they are small in size roughly 50-100mb and a lot of other objects such as distribution lists and contacts (thousands!!)).
My EXCH 2013 VM is configured with 4 vCPUs and 32GB of ram (according to VMWare this spec is one NUMA node).
My databases are configured for small size, each hosts about 10 users to keep the average database size at 100GB for backup and db maintenance efficiency, thus I have 40 databases, 20 live and 20 archive.
As I am batch migrating 50 users over I've noticed my CPU getting pegged and the migration is SLOW, i've looked at some Exchange blogs and used AnalyzeMoveRequestStats.ps1 from MS Exchange Team (http://blogs.technet.com/b/exchange/archive/2014/03/24/mailbox-migration-performance-analysis.aspx) to analyze where the slowness is and it reported as such..
Name : ProcessedStats1
StartTime :
EndTime : 7/28/2014 5:15:13 PM
MigrationDuration : 05:56:17
MailboxCount : 48
TotalGBTransferred : 36.24
PercentComplete : 61.48
MaxPerMoveTransferRateGBPerHour : 0.20
MinPerMoveTransferRateGBPerHour : 0.10
AvgPerMoveTransferRateGBPerHour : 0.14
MoveEfficiencyPercent : 22.76
AverageSourceLatency :
AverageDestinationLatency :
IdleDuration : 68.08 %
SourceSideDuration : 1.75 %
DestinationSideDuration : 3.94 %
WordBreakingDuration : 2.65 %
TransientFailureDurations : 0.00 %
OverallStallDurations : 92.39 %
ContentIndexingStalls : 0.00 %
HighAvailabilityStalls : 0.00 %
TargetCPUStalls : 92.39 %
SourceCPUStalls : 0.00 %
MailboxLockedStall : 0.00 %
ProxyUnknownStall : 0.00 %
We can see the efficiency is horrible and that the bottle neck is TargetCPUStalls (Exchange 2013 VM). Looking at the VM over the 6 hours the migration has been running it is a steady 80% used CPU.
My though is to increase the vCPUs from 4 to 8 and the RAM from 32GB to 64GB, will this be an overkill for 300 mailboxes with a fairly large average mailbox size (15-20gb/per mailbox)? This should make the VM 2 NUMA nodes according to VMWare (1 numa node = 4x vCPUs / 32GB RAM)..
My VM HOST is 2 Processor Sockets, 8 Cores and 32 Logical Processors (E5-2680) with 256GB RAM (3 Hosts in a VM Cluster).
Any expert suggestions are appreciated on this matter.
