VMworld 2007 Presentation with Todd Muirhead

1. Session S288993
Virtualizing Exchange
2007:
The Final Frontier?
September 2007
S t b
Todd Muirhead, Enterprise TechCenter, Dell
Kong Yang, Vi t li ti S l ti
K Y Virtualization Solution
Engineering, Dell

2. Some or all of the features in this document may be representative
of feature areas under development. Feature commitments must
not be included in contracts, purchase orders, or sales agreements
p g
of any kind. Technical feasibility and market demand will affect
final delivery.

3. Introduction
Microsoft Exchange 2007 Behavior On ESX S
Mi ft E h B h i O Server
Virtual Machines Are Not Well Understood
– How does Exchange 2007 p
g profile differ from Exchange
g
2003?
– 64-bit
– New server roles: Mailbox, Client Access, Hub Transport, Unified
Messaging, and Edge Transport
– Other new features such as CCR, administrative tools
– How will these features affect an Exchange 2007 VM?
– More memory utilization and features means that the
performance profile could change.
– The new server roles offer opportunities to virtualize the different
roles.
roles
– Where to start?

4. Stages of Virtualization
Adoption
Stage O
S One: Evaluation / Test and Dev
E l i T dD
– Introduction to virtualization
– Use of single server or small farm for test and
development
Stage Two: Virtualization of Some Production Level Apps
/ Consolidation
C lid i
– An initial set of “good candidates” for virtualization
– Farm of servers is used to support load-balancing high
load-balancing,
availability, and easy maintenance
Stage Three: Extend Virtualization
– Virtualize more applications to extend same advantages
as seen in earlier stages

5. Typical Application “Good
Candidates
Candidates” for Virtualization
Characteristics
Ch i i
– Low utilization
– Low I/O
– Running on old hardware
Examples
p
– Test and development servers
– Domain controllers
– DNS servers
– Intranet application servers
– Anything running on Pentium Pro servers!

6. Exchange 2007 –
The Final Frontier?
Exchange 2007 Is a Demanding Application
– 64-bit high I/O large RAM and large disk
64 bit, I/O, RAM,
– Not the first application that should be virtualized
– If Exchange 2007 can be virtualized, then other
demanding applications can b candidates as well
d di li ti be did t ll
Reasons to Virtualize Exchange 2007
– Want to leverage virtualization infrastructure already
in place for other apps
– Take advantage of increased administrative
flexibility
– Exchange 2007 server roles can be considered
separately
– Mailbox Hub Transport Client Access Edge Unified
Mailbox, Transport, Access, Edge,
Messaging

7. Exchange 2007 –
The Final Frontier?
Reasons not to Virtualize Exchange 2007
– Not officially supported by Microsoft1
– Easier-to-virtualize applications still exist in
environment
– Requirement for a very large number of users per
server
1 When considering the deployment of Microsoft Exchange Server 2003 on a virtualization platform such as
VMware ESX Server, please refer to support.microsoft.com/kb/897615 for information about the Microsoft
support policy for their applications running in non-Microsoft virtualization software.

8. Where to start with
Exchange 2007 VMs?
Capacity Planning and Reference Architecture
– Starting Point: Dell Physical Reference Architecture
for Microsoft Exchange 20071
– Evaluate four major subsystems: CPU, Memory,
Disk d Network
Di k and N t k
– Analyze Exchange 2007 VM Behavior under load
Exchange 2007 VM Behavior
– How will VMotion be affected?
– Can I run other VM types with Exchange 2007
VMs?
VM ?
– How does VMware HA perform with Exchange 2007
VMs?
1. www.dell.com/downloads/global/solutions/Dell_Exchange_2007_Advisor.pdf

9. Sizing Exchange 2007 VMs
g g
– Small, medium, and large VM configurations1
– Physical server specifications
– DellTM PowerEdgeTM 2950
– Two quad-core 2.33 GHz Intel® Xeon ®
– 16 GB RAM
– DELL / EMC CX3-80 – RAID 1/0 for data and logs
Small Medium Large
Users 500 1,000 2,000
RAM 8 GB 8 GB 16 GB
Data Disks 4 8 16
Log Disks 2 4 8
vCPU 1 to 4 1 to 4 1 to 4
vNIC 1 1 1
Storage Groups
g p 4 8 16
1. www.dell.com/downloads/global/solutions/Dell_Exchange_2007_Advisor.pdf

10. vCPU Scaling Test Results for
1VM @ 500 Users
– Results of 1, 2, 4 vCPU VM testing in small
configuration
– S dM il 95th response ti
SendMail time i b l
is below 560 ms.
– Overall CPU utilization is below 6% on an 8-core PE2950.
Microsoft Exchange 2007 Small VM Config
vCPU Scaling for 500 Heavy Users
580 6
560
5
ms)
540
SendMail 95th (m
CPU Utilization (%)
4
520
500 3
480
2
S
C
460
1
440
420 0
1vCPU 8GB 2vCPU 8GB 4vCPU 8GB
SendMail 95th Percentile (ms) 471 485 560
Host Server CPU Utilization (%) 3.5 4.5 5.3
VM Config
SendMail 95th Percentile (ms) Host Server CPU Utilization (%)

11. vCPU Scaling Test Results
for 1VM @ 1 000 Users
1,000
– Results of 1, 2, 4 vCPU VM testing in medium
configuration
– SendMail 95th response time is below 814 ms.
– Overall CPU utilization is below 13% on an 8-core PE2950.
Microsoft Exchange 2007 Medium VM Config
vCPU Scaling for 1000 Heavy Users
900 14
800
12
s)
700
endMail 95th (ms
CPU Utilization (%)
10
600
500 8
400 6
U
Se
300
4
200
2
100
0 0
1vCPU 8GB 2vCPU 8GB 4vCPU 8GB
SendMail
S dM il 95th P
Percentile
til 693 482 814
Host Server CPU Utilization 6.8 8.2 12.7
VM Config
SendMail 95th Percentile Host Server CPU Utilization

12. vCPU Scaling Test Results
for 1VM @ 2 000 Users
2,000
– Results of 1, 2, 4 vCPU VM testing in large configuration
– 1vCPU 16GB DNF test runs.
– 95th Response time is below 934 ms.
– Overall CPU utilization is below 24% on an 8-core PE2950.
Microsoft Exchange 2007 Large VM Config
vCPU Scaling for 2000 Heavy Users
1000 25
900
800 20
dMail 95th (ms)
CPU Utilization (%)
700
600 15
500
400 10
Send
300
200 5
100
0 0
1vCPU 16GB 2vCPU 16GB 4vCPU 16GB
SendMail 95th Percentile 934 664
Host Server CPU Utilization 12.5 22.3 23.4
VM Config
SendMail 95th Percentile Host Server CPU Utilization

13. Number of VMs
Testing Results
– How does performance of single large VM compare to
multiple smaller VMs in small configuration?
– 2 VMs with sufficient vCPU and memory resources performs best.
Microsoft Exchange 2007 VM Scaling
Small Config with 500 Heavy Users
900 12
800
10
ndMail 95th (ms)
700
CPU Utilization (%)
600 8
500
6
400
U
300 4
Sen
200
2
100
0 0
1VM 2vCPU 2VM 1vCPU 1VM 4vCPU 4VM 1vCPU 2VM 4vCPU 4VM 2vCPU
8GB (8GB 4GB (8GB 8GB (8GB 2GB (8GB 8GB (16GB 4GB (16GB
Total) Total) Total) Total) Total) Total)
VM Config
SendMail 95th Percentile Host Server CPU Utilization

14. Number of VMs
Testing Results
– How does performance of single large VM compare to
g g
multiple smaller VMs in medium configuration?
– 1 VM with sufficient vCPU and memory resources performs best.
Microsoft Exchange 2007 VM Scaling
Medium Config with 1000 Heavy Users
1400 20
18
1200
16
s)
CPU Utilization (%)
dMail 95th (m
1000 14
12
800
10
600
8
Send
U
400 6
4
200
2
0 0
1VM 2vCPU 2VM 1vCPU 1VM 4vCPU 4VM 1vCPU 2VM 4vCPU 4VM 2vCPU
8GB (8GB 4GB (8GB 8GB (8GB 2GB (8GB 8GB (16GB 4GB (16GB
Total) Total) Total) Total) Total) Total)
VM Config
SendMail 95th Percentile Host Server CPU Utilization

15. Number of VMs
Testing Results
– How does performance of single large VM compare to
multiple smaller VM i l
lti l ll VMs in large configuration?
fi ti ?
– 1 VM with sufficient vCPU and memory resources performs best.
Microsoft Exchange 2007 VM Scaling
Large Config with 2000 Heavy Users
5000 40
4500
35
4000
ndMail 95th (ms)
CPU Utilization (%)
30
3500
25
3000
2500 20
2000
15
U
Sen
1500
10
1000
500 5
0 0
1VM 2vCPU 2VM 1vCPU 1VM 4vCPU 4VM 1vCPU 2VM 4vCPU 4VM 2vCPU
16GB (16GB 8GB (16GB 16GB (16GB 4GB (16GB 8GB (16GB 4GB (16GB
Total) Total) Total) Total) Total) Total)
VM Config
SendMail 95th Percentile Host Server CPU Utilization

16. Heterogeneous VM Testing
with Exchange 2007 VMs
– What is the impact to different types of VMs running on the same server as a
Microsoft Exchange 2007 VM?
– Tested by running Exchange VM alongside SQL Server and SLES LAMP VMs.
– SQL and LAMP VMs overall performance decreased by 1-2%; Avg RT increased less
than 10ms.
Heterogeneous VMs running with Exchange 2007 VM
9000 60
8000
50
7000
e s)
sponse Tim (m
inute
6000 40
tions per M
5000
30
4000
Avg. Res
Operat
3000 20
2000
10
1000
0 0
8 SQL Server 2005 VMs 8 SQL Server 2005 VMs + 8 SLES LAMP VMs 8 SLES LAMP VMs +
Exchange 2007 VM with Exchange 2007 VM with
1000 users 1000 users
VM Configurations
Operations per Minute Avg RT (ms)

17. Heterogeneous VM Testing
with Exchange 2007 VMs
– Wh t i th i
What is the impact t th E h
t to the Exchange 2007 VM with other VMs running on host
ith th VM i h t
server?
– Exchange VM impacted more than the SQL and LAMP VMs.
– 95th percentile response time increased from 482 ms to 1211 ms.
– Exchange VM Contribution to host server CPU utilization: 8.2% alone increased to
19.7% CPU utilization with other VMs running.
1400
1200
1000
vg RT (ms)
800
600
Av
400
200
0
1 Medium 1000 User Exch VM 1 Medium 1000 User Exch VM + 8 SQL VMs + 8 LAMP
VMs
SendMail Avg RT (ms) SendMail 95th Percentile Avg RT (ms)

18. VMotion With Exchange
2007 VMs
– What happens to large heavily loaded VMs when moved?
– Complete 10 VMotions during a 2 hour period of Exchange 2007 activity
and record VMotion time and VM behavior
behavior.
– From Virtual Center, VMotion completes in all cases but Load Generator
shows some skipped tasks in the 1k user case and terminates in the 2k
case.
Small Medium Large Large
Users 500 1000 2000 2000
# of VMs 1 1 1 4
RAM per VM 16GB 16GB 16GB 4GB
vCPU 4 4 4 4
VMotion Times 3-5 Minutes 4-6 Minutes 10 Minutes 3-5 Minutes
Skipped Tasks No Yes N/A No
Task Queue > No No Yes No
5000

19. VMware High Availability
with Exchange 2007 VMs
–H
How d
does VM
VMware hi h availability f
high il bilit function with
ti ith
Exchange 2007 VMs?
– Testing VMware High Availability
– Remove power from host ESX server that has the Exchange 2007 VM
handling Load Generator tasks.
– VMware HA brings the VM online on another ESX Server.
– Observe and record the down time and the behavior after the VM
comes online.

20. Testing High Availability
with Exchange 2007 VMs
– Results from Testing VMware HA
g
– The Exchange 2007 VM had a downtime of a few minutes.
– 500 users had 563 exceptions; 1000 users had 1106 exceptions; 2000
users did not recover.
VMware HA Testing with Microsoft Exchange 2007 VMs
3500 1200
3000 1000
# of Tas Exceptions
s)
2500
SendM 95th (m
800
2000
600
ail
sk
M
1500
400
1000
500 200
0 0
500 Users 1000 Users 2000 Users
Configuration
SendMail 95th # of Task Exceptions

21. Summary of Observations
for Exchange 2007 VMs
– B t Practice for Exchange 2007 i a VM
Best P ti f E h in
– Utilize Dell Exchange Reference Architecture for sizing RAM,
storage, and network
– Th number of vCPUs per VM d
The b f CPU depends on th number of users
d the b f
to be supported
– Small – 1vCPU, medium – 2vCPU, large – 4vCPU
– Exchange 2007 VM does not greatly affect overall performance of
other VMs; but its overall performance was impacted.
– VMware HA was effective in bringing small, medium Exchange
2007 VMs back online from a simulated host server failure.
– VMotion of the Small Exchange 2007 VM was successful.
– Larger Exchange 2007 VMs require more investigation

22. Conclusions
– Mi
Microsoft E h
ft Exchange 2007 f ll capable i VI3
fully bl in
– Stacking multiple, small Exchange VMs improves
p
VMotion, HA capabilities
– Design Exchange 2007 VMs from physical reference
Architecture
– Ensures basic quality of service for storage and network subsystems
– Provides tuning of CPU/memory through resource allocation in VI3

23. Questions?
Session S288993
S i
Virtualizing Exchange 2007: The Final Frontier?
Todd Muirhead and Kong Yang
Dell Inc.
For more information …
www.dell.com/vmware
www.dell.com/techcenter
www.delltechcenter.com