Figure 14 – Single disk buffered overhead. Buffered overheads for both basic and dynamic volumes show little difference. 2KB writes show a significantly higher overhead than reads. This is due to the additional 4KB read and merge that takes place before each 2KB write, effectively more than doubling the request overhead incurred.
|
The right graph implies that for dynamic volumes, a single processor saturates at 200MBps, and two processors saturate at 400MBps given 5ms of processor time per MB with 64KB read requests. For writes, one and two processors saturate at 125MBps and 250MBps respectively given 8ms per MB written with 64KB requests.
|
