|
Figure 26 shows single disk unbuffered throughput across both a SCSI and an IDE disk. Like buffered throughput, unbuffered throughput on the Fireball IDE drive shows good sequential performance. At one deep requests, maximum disk throughput is attained at read requests of 16KB or more and write requests of 64KB or more. Optimal throughput is reached through two-deep reads and one-deep writes (except on the 3ware card which required two-deep writes) which both attain disk speed at 8KB requests. Increased depths showed no additional improvement in throughput.
Unlike the unbuffered SCSI throughput, IDE read throughput suffers at small 2KB request sizes. This is an artifact of the 3ware card and not of the Fireball drives. As Figure 26 shows, when the Fireball IDE drive is connected to the motherboard’s IDE channel using DMA, small requests show the same general trend lines as SCSI; 2KB read requests no longer have the dip seen with the 3ware card. The 3ware wasn’t handling as many IOs per second as the motherboard IDE channel.
|
