Baseline Performance for Mailbox Server (Using MAPI)
This section provides baseline performance data on Exchange Server 2003 mailbox server under specific MAPI client loads. The key hardware criteria for a mailbox server are processor type and speed, memory size, network speed, and disk performance and configuration.
Note
Exchange 2000 Server scalability data is used as a basis for comparison. Load Simulator 2000 and Load Simulator 2003 are used to simulate Outlook 2000 and Outlook 2003 MAPI clients respectively. For more information about Load Simulator (LoadSim), see Exchange Server 2003 Performance Tools.
Hardware
The following table shows the hardware specifications used in the following scenarios:
Scenario 1: Exchange 2000 Server and Outlook 2000 Online versus Exchange Server 2003 and Outlook 2003 Online
Scenario 2: Exchange 2000 Server and Outlook 2003 Cached Exchange Mode versus Exchange Server 2003 and Outlook 2003 Cached Exchange Mode
Mailbox hardware configuration
Server type | Processor type | RAM | Storage |
---|---|---|---|
Back-end server |
Intel P4 Xeon 4 processors, 1.4 GHz (Hyper-Threading disabled) |
4 GB |
|
Scenario 1: Exchange 2000 Server and Outlook 2000 Online vs. Exchange Server 2003 and Outlook 2003 Online
This scenario compares the load characteristics of Exchange 2000 Server and Outlook 2000 with Exchange Server 2003 and Outlook 2003. Load Simulator 2003 is used to generate the load. The following LoadSim configuration was used:
LoadSim Client Configuration
Intel P4 600 megahertz (MHz)
512 MB RAM
Single Integrated Drive Electronics (IDE) disk
1,000 LoadSim clients per computer
LoadSim Configuration The following tables show the values that were used.
Mailbox configuration settings
Mailbox configuration Value Messages in Inbox
100
Message in Deleted Items
1
Number of new folders
10
Messages per new folder
10
Calendar Appointments
25
Number of Contacts
128
Average Message Size
75 KB
User action settings
User actions per 8-hour day Value Send Mail
12
Messages Received/day
208
Messages Sent/day
52
Process Inbox
12
Browse Mail
20
Check Free/Busy
1
Request Meeting
1.4
Make Appointment
2.8
Browse Calendar
6
Journal Applications
0
Logoff
0
Browse Contacts
0
Create Contact
0
Distribution list settings
Distribution lists Value Distribution lists per site
100
Minimum distribution list size
2
Average distribution list size
10
Maximum distribution list size
20
The following table shows the results produced by the mailbox server in this scenario.
Mailbox server performance comparison
Server platform | Windows 2000 SP3 | Windows 2000 SP3 | Windows Server 2003 | Windows Server 2003 |
---|---|---|---|---|
Exchange platform |
Exchange 2000 Server SP3 |
Exchange Server 2003 |
Exchange Server 2003 |
Exchange Server 2003 |
Client platform |
Windows XP, Outlook 2000 Online |
Windows XP, Outlook 2003 Online |
Windows XP, Outlook 2003 Online |
Windows XP, Outlook 2003 Online |
Hyper-Threading enabled? |
No |
No |
No |
Yes |
Users |
8,000 |
8,000 |
8,000 |
8,000 |
% Processor Time |
75% |
76% |
75% |
57% |
Context Switches/sec |
8,300 |
7,833 |
10,183 |
13,487 |
Local Delivery Rate |
30 |
30 |
30 |
30 |
Network Usage (in Kbps) |
3,154 |
1,594 |
1,604 |
1,613 |
DB Disk Transfers/sec |
2,088 |
1,980 |
1,929 |
1,955 |
Log Writes/sec |
377 |
314 |
353 |
398 |
Disk Bytes/sec (MB) |
19.9 |
20.4 |
18.9 |
17.9 |
Database Cache Size |
864 |
896 |
896 |
896 |
RPC Operations/sec |
950 |
955 |
940 |
925 |
RPC Requests |
7 |
8 |
8 |
7 |
Client Latency (ms) |
96 |
100 |
95 |
105 |
Store Virtual Bytes |
2,047 |
2,075 |
2,080 |
2,082 |
For more information about the performance counters used in this scenario, see Performance Counter Definitions.
Processor
Exchange 2000 Server SP3 and Exchange Server 2003 show comparable levels of processor performance. The 8,000-user Exchange 2000 Server SP3 test reaches 75 percent processor usage, and the Exchange Server 2003 test reaches 76 percent processor usage. Both tests show similar workloads (950 RPC Operations/sec to 955, with the same Local Delivery Rate).
The table above shows that Windows 2000 Advanced Server with SP3 and Windows Server 2003, perform comparably in terms of processor usage when running Exchange Server 2003. However, Windows Server 2003 has key memory manager optimizations that significantly reduce virtual memory fragmentation, which is not shown in the table.
Exchange Server 2003 takes advantage of Intel Hyper-Threading Technology to increase server scalability by 25 percent. This table shows that processor use is reduced by 25 percent (from 76 percent to 57 percent) when the same test is run with Hyper-Threading enabled. The benefit of Hyper-Threading is the same for Cached Exchange Mode client scenarios.
Memory
The Exchange store memory footprint of Exchange Server 2003 is approximately the same as that of Exchange 2000 Server. With the increase in Database Cache Size (896 MB compared to 864 MB) in Exchange Server 2003 and performance improvements in Jet, random Exchange store disk I/O operations can be reduced by up to 10 percent.
Disk Usage
Exchange Server 2003 produces 10 percent fewer Exchange database disk I/O operations than Exchange 2000 Server SP3. Because of the increase in Database Cache Size (896 MB compared to 864 MB) in Exchange Server 2003, random Exchange database disk I/O operations can be reduced by up to 10 percent.
Network Usage
Exchange Server 2003 and Outlook 2003 can reduce network bandwidth when compared to Exchange 2000 Server and Outlook 2000 by up to 50 percent. This 8,000-user test shows that Exchange 2000 Server and Outlook 2000 use 3,154 Kbps, and Exchange Server 2003 and Outlook 2003 use approximately half the bandwidth at 1,594 Kbps. Because of the compression technology implemented in Exchange Server 2003 and Outlook 2003, the network bandwidth required to host a particular number of users is significantly less than that of Exchange 2000 Server and Outlook 2000. Many performance optimizations were implemented in Exchange Server 2003 so this compression would not detract from the scalability of the server. In effect, the reduction of bytes over the wire is free in terms of server CPU usage.
Scenario 2: Exchange 2000 Server and Outlook 2003 Cached Exchange Mode vs. Exchange Server 2003 and Outlook 2003 Cached Exchange Mode
This test compares the load characteristics of Exchange 2000 Server and Outlook 2003 with Exchange Server 2003 and Outlook 2003 in a Cached Exchange Mode client scenario. LoadSim 2003 is used to generate the load.
LoadSim Client Configuration
Intel P4 600 MHz
512 MB RAM
Single IDE disk
1,000 LoadSim clients per computer
LoadSim Configuration The following tables show the values that were used.
Mailbox configuration settings
Mailbox configuration Value Messages in Inbox
100
Message in Deleted Items
1
Number of new folders
10
Messages per new folder
10
Calendar Appointments
25
Number of Contacts
128
Average Message Size
75 KB
User action settings
User actions per eight-hour day Value Send Mail
7
Messages Received/day
161.90
Messages Sent/day
44.12
Process Inbox
20
Browse Mail
0
Check Free/Busy
4
Request Meeting
2
Make Appointment
4
Browse Calendar
1
Journal Applications
0
Logoff
3
Browse Contacts
0
Create Contact
0
User action settings
Distribution lists Value Distribution lists per site
100
Minimum distribution lists size
2
Average distribution lists size
10
Maximum distribution lists size
20
The following table shows the results produced by the mailbox server in this scenario.
Exchange 2000 Server/Exchange Server 2003 Cached Mode comparison
Server platform | Windows 2000 SP3 | Windows 2000 SP3 |
---|---|---|
Exchange platform |
Exchange 2000 Server SP3 |
Exchange Server 2003 |
Client platform |
Windows XP, Outlook 2003 Cached Exchange Mode |
Windows XP, Outlook 2003 Cached Exchange Mode |
Hyper-Threading enabled? |
No |
No |
Users |
8,000 |
8,000 |
% Processor Time |
64% |
57% |
Context Switches/sec |
7,981 |
6,488 |
Local Delivery Rate |
23 |
23 |
Network Usage (in Kbps) |
3,613 |
1,765 |
DB Disk Transfers/sec |
1,496 |
1,355 |
Log Writes/sec |
304 |
315 |
Disk Bytes/sec (MB) |
17.7 |
16.2 |
Database Cache Size |
896 |
864 |
RPC Operations/sec |
1,205 |
1,193 |
RPC Requests |
8 |
7 |
Store Virtual Bytes |
2,016 |
2,076 |
For more information about the performance counters used in this scenario, see Performance Counter Definitions.
Note
Results of the Outlook 2003 online tests and the Outlook 2003 Cached Exchange Mode tests are not comparable. The user profile and user actions are quite different. For specific performance data about Outlook 2003, see Client Network Traffic with Exchange Server 2003.
Processor
In this test, processor performance of Cached Exchange Mode MAPI clients on Exchange Server 2003 is approximately 10 percent better than Exchange 2000 Server SP3. The 8,000-user Exchange 2000 Server SP3 test reaches 64 percent processor usage, and the Exchange Server 2003 test reaches 57 percent processor usage. Both tests show similar workloads (1,193 RPC Operations/sec compared to 1,205, with the same Local Delivery Rate). Exchange 2003 Server is optimized for Outlook 2003 Cached Exchange Mode clients.
Memory
The Exchange store memory footprint of Exchange Server 2003 is approximately the same as that of Exchange 2000 Server. With the increase in Database Cache Size (896 MB compared to 864 MB) on Exchange Server 2003 and performance improvements in Jet, the random Exchange database disk I/O operations can be reduced by up to 10 percent.
Disk Usage
Exchange 2003 produces 10 percent fewer Exchange database disk I/O operations than Exchange 2000 Server SP3. Because of the increase in the size of the database cache (896 MB compared to 864 MB) in Exchange Server 2003, the random Exchange database disk I/O operations can be reduced by up to 10 percent.
Network Usage
Exchange Server 2003 and Outlook 2003 can reduce the network bandwidth when compared to Exchange 2000 Server and Outlook 2003 by up to 50 percent. This 8,000-user test shows Exchange 2000 Server and Outlook 2003 uses 3,613 Kbps, and the Exchange Server 2003 and Outlook 2003 test uses approximately half the bandwidth at 1,765 Kbps. Because of the compression technology implemented in Exchange Server 2003 and Outlook 2003, the network bandwidth required to host a particular number of users is significantly less than that of Exchange 2000 Server and Outlook 2003. Many performance optimizations were implemented in Exchange Server 2003 so this compression would not detract from the scalability of the server. In effect, the reduction of bytes over the wire is free in terms of server CPU usage.
Mailbox Scalability Guidelines
When you design a mailbox server, consider the following recommendations.
A back-end mailbox server hosting many MAPI connections scales well on 4-processor servers.
Generally, one processor for every 1,000 users is a good guideline for all but the heaviest of mail users. For more information on processor requirements, see Calculate Your Server Size.
Exchange Server uses a maximum of 3 GB of memory. To increase performance, increase memory up to 4 GB to reduce the paging to disk.