Under Pressure (da da dum diddy dum dum)...

First off, welcome to the new subscribers who hopped on-board with BeingExchanged!  Also, congratulations to everyone who recognized the riff I used in my Subject line, and for those who don't, it's referring to This Song, and don't we all feel old?

Back Pressure is a phenomenon in Exchange 2007 that can grind your email flow to a halt with little or no warning, so it's a great thing to begin to understand now, before you run into it in the real world.  The basic idea is that Exchange Servers through the years were famous (sorry in advance to the guys from Microsoft) for collapsing under heavy I/O load and/or lack of disk space.  Back Pressure allows the Exchange 2007 system to halt some of the processing of data temporarily to allow the system to keep running without the danger of I/O overload.

Specifically, Back Pressure allows Exchange Servers to stop accepting new messages into transport queues, thereby allowing system resources to free up over time and for those queues to empty out.  This translates into no mail coming or going from your organization while Back Pressure is in effect, which will have quite an impact on your operations.

While the Back Pressure system could impact your mail organization as a whole, it's good to point out that only Hub/Transport and Edge Services servers are impacted.  Messaging queues and the disks they sit on are what the Back Pressure systems monitor, not database locations.  Pure Mailbox Role (MBX) servers won't experience Back Pressure, but they will be impacted by it if mail cannot flow in or out of the MBX servers due to mail being unable to transverse the other roles.

When the Back Pressure system kicks into gear, the most noticeable impact is that Outlook clients will have all their outgoing mail stuck in either the Outbox or Drafts folders.  OWA will show everything in Drafts and other client software will show that mail isn't being sent.  Of course, no mail is coming in either, but that's a little more difficult to "see" unless the user was expected a message.  On further examination, you will see that no mail is coming into or going out of mail queues on the Hub Transport or Edge servers, a dead giveaway that Back Pressure kicked in if your connectivity is working just fine but no mail is moving.

There are three ways Back Pressure can be brought online on an H/T or Edge server:

First off, if the logical disk that contains logging and other information for either role runs below a set level of available space, Back Pressure will be invoked.  This is done to keep Exchange from "crashing" a disk, or filling up all space available and causing problems within the Exchange system or (if it's also the system drive) Windows itself.

Next, if the amount of physical memory or other system resources run low; Back Pressure comes online.  Since mail isn't coming in or going out, the resources will return to a normal level once whatever process is spiking finishes or is terminated.

Third, the number of pending operations in the memory queue (as opposed to the overall amount of memory in use) can build up if something is slowing down Exchange like a virus, DDOS attack or just a sudden influx of mail transfer.  If the depth of these transactions exceeds a limit set within Exchange, the result is Back Pressure stopping mail flow, allowing the existing transactions in memory to commit through to disk without new transactions coming in.

Each of these metrics has a different number of resources assigned as the "break point" where Back Pressure kicks in.

For disks, Exchange uses a formula to determine what space must be present and free before Back Pressure is invoked.  Specifically the formula is 100*(hard disk drive size - fixed constant) / hard disk drive size. The "fixed constant" number is 4GB for the release version of Exchange 2007, 500MB for the SP1 version. If the total available free space on the drives which contain queuing systems drops below that number, new mail flow stops and Exchange system spools out the queued data until the disk systems free up enough space to get above the threshold.

Finding the amount of memory used to reach the "high" utilization number, or the depth of the queues is a bit trickier.  There are several calculations and settings used to figure those out, and a complete description can be found in this article about Back Pressure technology in Exchange 2007.

The end result is the same.  Build an Exchange 2007 server with the wrong sized resources (disks, memory or server load), and the system will effectively halt itself to make sure that those limited resources don't crash.  So it is vital to properly size out both your expected user load, and the resources (physical and otherwise) that you're planning on running that load on.  You can change the thresholds for the Back Pressure triggers (see this link) but that's designed to be more of a temporary fix than a permanent change.  Instead, work with the publicly available sizing guides from Microsoft and other sources to make sure your hardware can handle the mail transfer without feeling the pressure.