Documente Academic
Documente Profesional
Documente Cultură
Disclaimer
2012 Microsoft Corporation. All rights reserved. Microsoft, Office 365, and other product and service names are or may be registered trademarks and/or trademarks in the U.S. and/or other countries. The information herein is for informational purposes only and represents the current view of Microsoft Corporation as of the date of this presentation. Because Microsoft must respond to changing market conditions, it should not be interpreted to be a commitment on the part of Microsoft, and Microsoft cannot guarantee the accuracy of any information provided after the date of this presentation. MICROSOFT MAKES NO WARRANTIES, EXPRESS, IMPLIED OR STATUTORY, AS TO THE INFORMATION IN THIS PRESENTATION.
Performance testing and validation is currently in progress on Exchange 2013 the Exchange Product Group has not yet finalized their guidance on virtualizing E2013.
Guidance may change, keep an eye on TechNet for the latest on virtualization guidance
About me
Senior Consultant with Microsoft Consulting Services At Microsoft for nearly five years and counting Instructor for Microsoft Certified Master | Exchange
Virtualization Load Balancing RBAC
Underutilized processors == too much money spent Underutilized memory == too much money spent
Politics
We bought it, were going to use it everywhere CIOs / managers reading CIO Magazine again Senior company leadership
SUPPORTABILITY
E2010 RTM: All roles except UM E2010 SP1+: All roles including UM
- Differencing/delta/dynamic disks
- Not supported for any Exchange role
Hypervisor snapshots make lab testing much easier Resist the temptation to use hypervisor snapshots in production they arent supported! Aspects of the Exchange system do not handle time travel well (notably log shipping) Use caution with snapshots in the lab (you may need to roll back more than one VM)
Planning Guidance
Historically, our planning guidance has been very simple: Determine your baseline requirements, then have the physical/virtual discussion. Why?
You still need enough CPU You still need enough memory You still need enough disk I/O You still need enough network bandwidth
Requires Exchange 2010 SP1+ for support VM should have a minimum of:
(4) Virtual Processors/Cores Requires 1:1 vCore:pCore allocation 16GB of memory
Do not deploy MBX VMs in the same DAG on the same root server
HUB
MBX
VS
HUB HUB
MBX
Host-Based High Availability What is Host-Based High Availability? Automatic failover of virtual machines to another virtualization host in the event of a critical hardware failure (virtualization platform independent)
Host Based High Availability What you need to be aware of: Not an Exchange-aware solution
No knowledge of transaction logs, clean/dirty database dismount, database checksums, Exchange-health, etc.
Only protects against server hardware/network failure No protection against storage failure / data corruption Trend is larger mailboxes = larger database sizes = longer time to fully recover from data loss (cache warming) Requires the guest VM to perform a cold boot (this is HA?)
Live Migration / vMotion / etc. What is Not Supported: Quick Migration (Windows Server 2008 pre-R2) Anything that pauses/saves state, migrates, and then resumes Time Traveling
Enable jumbo frames on the Live Migration network(s) Use very fast networks (5Gb, 10Gb)
Which HA to choose?
Make sure your DAG members are placed where you think they should be, and that they dont migrate where they shouldnt be
Interesting DR scenarios
Some physical, some virtual Ensure that if you have a virtual machine as a target for multiple physical server in your DR scenario, that it can handle the worst case DR
Test as a system
Jetstress and LoadGen all VMs on the system to stress the entire system with a realistic peak load Just like with physical, one or the other is not enough Unless your clients are going to be virtual in production (VMware VDI for instance), test with physical LoadGen clients
Network
Bandwidth Take into account iSCSI when necessary!
High bandwidth customers should avoid 1Gb iSCSI Newer FC4Gb, FC8Gb, 10Gb iSCSI or FCoE better
Deployment Recommendations Exchange Server 2010/2013 is not virtualization aware Core Exchange Design Principles Still Apply
Design for Performance, Reliability and Capacity Design for Usage Profiles (CAS/MBX) Design for Message Profiles (Hub/Edge)
100
150 200 250 300 350 400
6
9 12 15 18 21 24
450
500
27
30
50
0.15
1.1
0.17
100
150 200 250
2
3 4 5
0.3
0.45 0.6 0.75
2.2
3.3 4.4 5.5
0.33
0.5 0.66 0.83
300
350 400 450
6
7 8 9
0.9
1.05 1.2 1.35
6.6
7.7 8.8 9.9
0.99
1.2 1.32 1.49
500
10
1.5
11
1.7
Unfortunately, too early guidance coming soon, watch TechNet and EHLO Expect that the 10% rule of thumb will continue
iSCSI
Standard best practice for iSCSI connected storage apply (dedicated NIC, jumbo frames, offload, etc.) iSCSI initiator in the guest is supported but need to account for reduced performance
Clarify support between hypervisor vendor and storage vendor (certain combinations are not supported)
The storage calculator but much, much more now! The de facto Exchange calculator Created during Exchange 2007 product cycle Properly calculates host CPU, memory, storage requirements, network bandwidth for replication Newer versions generate database/dag deployment scripts
This allows you to put in a processor and number of cores, and have the tool give you the average SpecInt value you can use in the calculator
Latest Mailbox Server Role Requirements Calculator (v17.2) now includes support for virtualization!
Hypervisor CPU Adjustment Factor is the virtualization overhead (rule of thumb is 10%)
You also have to manage that virtualization platform Are there alternatives? Do they make sense?
Can a physical architecture make full use of hardware? Would use of blade server technologies make more sense than virtualization?
Current Microsoft guidance is to deploy multi-role servers (CAS, HUB, MBX) instead of virtualization Your core math works out:
Example 8-core Physical Machine Mailbox Server Role gets 4 cores Client Access Server Role gets 3 cores (3:4 sizing) Hub Transport gets 1 core (1:5 sizing, this is better)
Additional reading
Windows Server 2008 R2 Hyper-V
http://www.microsoft.com/windowsserver2008/en/us/hyperv-main.aspx
Thank you!