SQL 2000 Low Perfomance

[Pieter 't Hart]

see ftp://ftp.compaq.com/pub/supportinformation/papers/tc020604tb_rev0_us.pdf

The performance of a RAID array can be subdivided into read performance and write performance, which vary based on the RAID level. For example, in RAID 1 (mirroring), the data and redundant data must be written to two separate drives. The data is normally read from one drive, so the read performance is much faster than the write performance; however, data can be read from both drives to increase the read performance. RAID 0 uses striping to improve performance by splitting up files into smaller portions (stripes) and distributing them to multiple hard disks; however, RAID 0 has no fault tolerance. RAID 5 and RAID ADG also use striping, but their write performance is significantly affected by the multiple reads and writes needed to perform the parity calculations prior to updating the array. The write performance of RAID ADG is less than that of RAID 5 because RAID ADG has dual parity overhead. The read performance of RAID 5 and RAID ADG is very good and may be improved by tweaking the stripe size.

[Pieter 't Hart]

Thats where the writeback cache comes in.
When data is written to cache. as for the OS the i/o is finished. The raid controller still has to perform the write to disk(s) in the background, but the OS (database) can submit a new write operation.

To guard for power failures before the cache is written to disk it has it's own battery. So when power is restored to the controller data can still be read from cache and written to physical disk.

A database heavily relies on "knowing" the certainty that the transactionlog record is written to "disk", but it sees no difference between disk and cache.