Transaction Logging can represent the single saving grace and at the same time the largest bottleneck in any version of Exchange 5.5, 2000 and 2003. In this article I will explore the reasons and mechanisms behind that and look at one of the reasons why your Exchange server could be running slow.

In order to understand what transactional logging is, let imagine you’re using an ATM. You’d start by typing in your secret code, after validation, you would choose an amount and an account to debit. Then the money counter would start spinning, the dispenser would be just about to open and – nothing, screen goes blank, dispenser stays shut. Power Failure.

What happened to your money you were about to withdraw? Do you hang around waiting for the power to come back on so that you can get your money? Does the next person in line get your money? What happens now?

Thank goodness bank ATM’s support something called Transactional logging. This means that every operation is written to disk for later comparison and action. In real life, when the ATM switches on, it reviews its logs, knows there was a withdrawal of money from a bank account in progress, notices the money has not been taken from the cash dispenser, and ROLLS BACK the transaction, restoring your bank account to it’s previous state. If it wasn’t for that you would be poorer and the next guy withdrawing money might get your previous amount.

How does this relate to Exchange? Exchange databases support transactional logging. This means that if you lost your entire database, but had every single transaction log, you could conceivably rebuild your Exchange mail store to the point where you lost it. This also means that if your Database shuts down in a dirty or unknown state, you may recover it to a known good state and remount it, after fixing the reason for the corruption.

Transaction logs normally offer this level of protection provided they don’t co-exist on the same disk as the transactional database and aren’t set to circular logging, otherwise a disk failure would amount to total loss. Logs tend to have at least one dedicated spindle (disk) or set of spindles (disk stripe - perhaps in a RAID 0+1 configuration) per transaction log.