I'm sizing a new storage system for a linux mailserver. Current system is about 8 years old, and working fine, but spares are becoming an issue - SCSI disks are increasingly harder to find, and our storage array manufacturer has been out of business for 3 years now, so I need to upgrade to more modern hardware.
The new system needs to support the same workload as the current system while allowing some room for growth. My plan was to get a sample of the current workload using iostat and then size a system to accommodate this. My (limited) understanding is that there's two main areas of performance - IOPS and data throughput (MB/s).
- Am I correct in assuming that "r/s" and "w/s" from iostat are analogous to IOPS, and "rKb/s" and "wKb/s" are analogous to data throughput?
- I've got two different sets of data from iostat - one with data points at one-second intervals and one with data points at one-minute intervals. Obviously the data points on a one-minute average are lower than those on a one-second average. Should I base my analysis on the one-second intervals? The data at one-second intervals is very bursty - periods of intense activity surrounded by periods of relative inactivity. If I size for the bursts, then I'm afraid I might be going overboard. Is there another interval I should be looking at, or another way of interpreting this data?
- Any other helpful tools or info you can point me towards?
TL,DR: I need a storage guru to beat me with a cluebat.
[link] [7 comments]