Desktop vs. Server Access - Localization Matters

The graph above lays out the access pattern of an "intense" single-user session. An average queue depth of 4 was achieved by copying a folder of files (9.31 GB, 2035 files of varying sizes) to another drive while copying the same set of files from the other drive to this target drive while defragmenting the disk while running a full spyware scan while running a full antivirus scan.

Note that even under this highly unlikely concurrency, accesses remain localized; in other words, the drive attempts to get as much done within a given area before surrendering itself to a broad stroke that takes its read/write heads to a different location. As a result, a drive's read-ahead and write-back strategies associated with the buffer and firmware code weigh heavily on overall performance.

Thus, manufacturers spend considerable time tweaking firmware/code on a drive intended for the single-user market.

The graph above represents the truly random access generated by iometer with an average queue depth of 4. Dispersed reads and writes such as these permeate large database and OLTP servers. Note that since no localization is present, any time spent reading ahead (or behind) for additional data will usually result in performance hit.

As a result, effort spent tweaking firmware/code for single-user access patterns on a drive intended for the enterprise market is wasted and may even be detrimental.