Client Network Traffic with Exchange 2007


EDIT 5/30/2008: The content of this article has since been published in our official Exchange documentation. Please go here for the most updated version.

As part of a broader investigation, we measured the network costs between Enterprise email clients and Exchange 2007 SP1. The values presented here might help you get a ball-park value for the network requirements connecting your datacenter to your users. The clients we considered were: Outlook 2007 Online mode; Outlook 2007 Cached mode; Outlook 2007 Cached mode via RPC/Http (Outlook Anywhere) and Outlook Web Access. We are not reporting the network bytes passed between Exchange roles, but the bytes entering and leaving the 'datacenter.' Outlook Anywhere and Outlook Web Access connect to the 'Client Access Server' role, while Outlook 2007 (in both Online and Cached mode) connects directly to the 'Mailbox' role. The network traffic from earlier Outlook versions can be estimated from the Exchange 2003 results, published in the whitepaper http://www.microsoft.com/technet/prodtechnol/exchange/2003/clinettraf.mspx, as there haven't been fundamental changes in the Exchange-Outlook communications in the 2007 releases.

Our user profile started with the message send and delivery rates from the 'light, medium, heavy and very heavy' knowledge worker profiles. We further assumed: an average message size of 50kB; that every message delivered was read; and half of the incoming mail was deleted. Our Web client logged in - logged off twice/day, while we neglected the logon - logoff costs from the other clients. (Enterprise email users tend to stay logged in for days at a time.)

Profile

Light

Medium

Heavy

Very Heavy

Sent /day

5

10

20

30

Received /day

20

40

80

120

Ave message Size

50k

50k

50k

50k

Messages read /day

20

40

80

120

Messages deleted /day

10

20

40

60

OWA logon-logoff /day

2

2

2

2

The network bytes transferred for each action is independent of mailbox size, so we did not perform separate measurements for each profile, but measured the costs of the actions and summed them for each profile.

Note that for Outlook 2007 in Cached mode and Outlook Anywhere, which work off a local copy of the user mailbox, there is negligible traffic associated with reading or deleting mail - as these actions work against the local copy - but every mail received is downloaded to the client.

In the table below, all values are in Kilobytes/day/user. I've broken out the sending portion from the other actions (labeled as 'aggregate').

Profile

 

Light

Medium

Heavy

Very Heavy

           
 

 Sending

190

380

760

1,140

Outlook 2007 -?? Online

aggregate

2,510

5,030

10,050

15,070

 

 total

2,700

5,410

10,810

16,210

           
 

 Sending

260

520

1,040

1,560

Outlook 2007 - Cached mode

aggregate

1,040

2,080

4,160

6,240

 

 total

1,300

2,600

5,200

7,800

           
 

 Sending

310

620

1,230

1,850

Outlook Anywhere 2007

aggregate

1,230

2,470

4,940

7,400

 

 total

1,540

3,090

6,170

9,250

           
 

 Sending

800

1,600

3,200

4,800

Outlook Web Access

aggregate

5,390

10,620

21,070

31,530

 

 total

6,190

12,220

24,270

36,330

To use these values, suppose you had a Datacenter with 10,000 Medium Outlook Cached mode users. Further assume all these users were in the same time zone, so they did the majority of the work during an 8 hour day.

This would predict the Average network bytes/sec would be

or

Assuming a daily peak of twice this average value, the network coming into the 'datacenter' would have to support roughly 15 megabits/sec from these users alone.

- Greg Smith


Share this post :

Comments (13)
  1. Joel Raper says:

    Good Post! Given your example, it would be good to get the needed Mb/Sec calc with Riverbeds or other WAN accelerators.  Maybe some folks out there have some real world examples.  This topic comes up a lot when talking with folks hoping to move towards centralization.  

    Thanks for the post!

  2. sarah says:

    Great insight!

  3. Ed says:

    Has anyone got figures for Outllok 2003 for similiar scenarios.

    Thanks

    Ed

  4. Andrew says:

    *Very* useful info THANK YOU!

    In related discussions I heard that a representative of a large California-based networking company alledged that "bandwidth usage between Outlook and Exchange has increased threefold with Exchange 2007".

    That sounds extremely improbable to me (as you point out "…there haven’t been fundamental changes in the Exchange-Outlook communications in the 2007 releases").

    Are you able to confirm that these allegations are unfounded?

  5. Jon says:

    What are they using blackberry?

  6. Derick says:

    Appreciate it if you could provide the answers to the following:

    1. How would I calculate the network bandwidth requirement per user using the calculation above? Can we assume the following below?

    For example for a medium profile Outlook cached user:

    = 1 X 2,600 / 8 X 3600

    = 0.09 KB/s

    = 0.72 kbps

    2. If the datacenter hosts users with different time zones, how will the calculation differ?

  7. Richard says:

    Did this testing use the light or premium versions of Outlook Web Access? What is the network bandwidth differences between the light and premium versions when testing with these user profiles?

  8. Greg Lambert says:

    Andrew — a factor of 3 increase seems unreasonable, but I can’t put my hands on data to quantify it.

    Derick — your calculation is correct for one user. Users spread out in different time zone will tend to even out the load, so you wouldn’t need as large a ‘peak factor’  (I used a factor of 2 in the last paragraph — based on the peak to average ratio of Microsoft’s corporate email flow).

    Richard — These were simulating OWA premium users. OWA light is optimized for low bandwidth networks by reducing the round trips (no spell check, for example) and a simpler UI (smaller gifs and less Jscript) — but the network bytes sent aren’t dramatically different from the premium client (which is reasonable, since the same message bodies are moved in both cases).

  9. Garry Trinder says:

    In your calculation, there is no mention of the heavy online https traffic (300k- 2Mb depending if you find a date available for 5-10 invitees) generated by the Scheduling Assistant in Outlook 2007; today, users are not only sending & receiving mails, but also book meetings…

  10. Nicholas Gimbrone says:

    You defined an "ave message size" as 50K, but have not defined what that represents. Is it for instance: (a) the body size, and if so, what form (plain text, html, rtf, etc); (b) the size of an SMTP representation of the entire "message" (e.g. headers, multi-part body as HTML & plain text and attachments); (c) the size of the sum of the sizes of all properties for the item (e.g. space consumed in the DB); (d) the size of the MSG file which Outlook would create if this item was saved to disk; (e) something else?

  11. Ed Crowley says:

    Does the Outlook Anywhere numbers also assume cached mode is employed at the same time?

  12. Wayne Hoggett says:

    I assume the figures would be similar for Outlook 2003?

    Ted makes a good point. Our Exchange environment is fast when sending and receiving emails, but viewing mulitple calendars and scheduling resources is slower than I would like.

    For those admins in the real world, remember DNS, AD Replication etc when calculating required bandwidth.

  13. Mindaugas says:

    What about SCCP 2007 and Exchange 2007 model? What formula is used in capacity planner Exchange model? When I put nubers from above example with "Average" profile and 10,000 users with OL2007 cached scenario, I got total (down+up) wan load of ~3Mbps. I suspect that OWA logon/logoff 2 times per day can increase the the bandwidth, but can it be 5 times more?

    Here are my settings:

    Mailbox site: Everything goes by default (1 local client, OL2007 cached, Average usage 100%)

    Client only site: Local client count=10,000, everything goes by default (1 local client, OL2007 cached, Average usage 100%)

    Bandwidth: site to site 11 Mbps, % Available – 100%

    Using default "Average" usage model:10 sent, 40 received, 50K size, peak factor – 2

    Results: Wan utilization Downlink – 26,73%, Uplink – 1,01%, so Total – 27,74%

    So if we take 11Mbps link and 100% available, consumed traffic would be 11*0,2774=3,05 Mbps (including peak factor)

    This value is 5 times less than 15Mbps.

    So  who is right?

Comments are closed.

Skip to main content