Best Practices for Maximizing Disk I/O

A primary rule of thumb is to separate sequential and random I/O. An Exchange 2000 Server database consists of multiple file types and access patterns, depending on the type of clients that your implementation supports. Messaging API (MAPI) protocol clients don't utilize the streaming store; for Internet protocol clients (i.e., IMAP, POP3, SMTP), the streaming store is the primary storage location. Also, each Exchange Server storage group (SG) has one set of shared database transaction logs, and you can configure each SG with multiple database files (i.e., combined property store—.edb—and streaming store—.stm—files). Because the Exchange Server database engine accesses the transaction log (.log) files in a strictly sequential manner, each SG should have a dedicated disk volume (preferably RAID 1 or RAID 0+1) in which to store the files. Depending on the supported clients, you might also choose to separate the property store and the streaming store on separate physical volumes. However, because of the cost of such a configuration and a lack of performance data justifying separation, most deployments choose to place both stores on the same volume and configure that volume as RAID 5 or RAID 0+1 for maximum performance and data protection. Table A shows some general storage guidelines.