Dedicated to VM and file system developers and researches...
It is immediately apparent from these figures that moving-arm disks should never be used, neither for paging applications nor for any other heavy-traffic auxiliary memory applications.
I wonder why disks don't have more than one "moving arm". With native command queue SATA disks (and something similar surely existing in the SCSI world for a long time), you could get quite an improvement in seek time and throughput... Or not actually?
I wonder why disks don't have more than one "moving arm" In fact, some of them do. High-end IBM installations use "fixed-head disks", where each track has its own head. You may imaging the cost of it.
High-end IBM installations use "fixed-head disks", where each track has its own head.
Well, this sound almost like the good old drum. And no, I really don't want to imagine the price of such a beast. ;-)
I was actually thinking of just sticking an arm into each corner of the case of your average disk: four arms, quarter the average seek time, and up to four times the throughput (yeah, I know this is a misleadingly coarse upper bound). And it doesn't seem to be that much more arms and heads to make it cost insanely more. Though, frankly, I have no idea what are the expensive parts of a disk, and making a firmware to use those four arms really well may also be a bit tricky.
Good luck finding a drum storage unit these days ;-)
ReplyDeleteI wonder why disks don't have more than one "moving arm". With native command queue SATA disks (and something similar surely existing in the SCSI world for a long time), you could get quite an improvement in seek time and throughput... Or not actually?
ReplyDeleteI wonder why disks don't have more than one "moving arm"
ReplyDeleteIn fact, some of them do. High-end IBM installations use "fixed-head disks", where each track has its own head. You may imaging the cost of it.
High-end IBM installations use "fixed-head disks", where each track has its own head.
ReplyDeleteWell, this sound almost like the good old drum. And no, I really don't want to imagine the price of such a beast. ;-)
I was actually thinking of just sticking an arm into each corner of the case of your average disk: four arms, quarter the average seek time, and up to four times the throughput (yeah, I know this is a misleadingly coarse upper bound). And it doesn't seem to be that much more arms and heads to make it cost insanely more. Though, frankly, I have no idea what are the expensive parts of a disk, and making a firmware to use those four arms really well may also be a bit tricky.
-- crankyjack