A. Anand, S. Sen, A. Krioukov, F. Popovici, A. Akella, A. Arpaci-Dusseau, R. Arpaci-Dusseau and S. Banerjee, "Avoiding File System Micromanagement with Range Writes", OSDI'08
Please provide an authoritative link to the full text of the paper.
Please also make the format consistent to the convention of this forum.
Range Writes separate block placement from file system which doesn't have enough
information and achieves better write performance.
From a typical disk such as :
http://www.seagate.com/ww/v/index.jsp?vg...http://www.seagate.com/ww/v/index.jsp?vgnextoid=c89ef141e7f43110VgnVCM100000f5ee0a0aRCRD#tTabContentSpec
The performance parameters are as follows:
Spin Speed (RPM) 7,200 RPM
Average latency 4.16ms
Random read seek time <8.5ms
Random write seek time <9.5ms
For a 7200RPM disk, the largest rotation latency is 60s/7200 = 8ms. Let's say
the expected average is 4ms (4.16ms given by the inventor). And the seek time
is about double of it.
From the results, we can see Range Writes can dramatically cut down the
rotation latency, but not so successfully optimize the seek time.
But Range Write requires changes on disk scheduling which is hard for OS. To
achieve the best performance, it's better to put disk scheduling on the
controller on disk. However, this requires change to the disk hardware.