1

So it seems to me that when trying to improve performance of a external memory device (HDD), the thing that is increased is RPM. Why not have multiple read/write heads per substrate instead of just increasing the RPM? Is this due to the formatting of the data or simply because heads are expensive?

SPQR
  • 11
  • 1
  • They used to... –  Sep 24 '20 at 16:26
  • 1
    Does [THIS](https://superuser.com/questions/1137805/why-arent-there-multiple-heads-covering-the-radius-of-a-hard-disk-platter) answer your question? – evildemonic Sep 24 '20 at 16:28
  • That link is good. Also note that stacking multiple disks and using both side of each does the same thing. – Transistor Sep 24 '20 at 16:30
  • ... see how small they were http://s3data.computerhistory.org/brochures/burroughs.onlinedisk.1962.102646217.pdf –  Sep 24 '20 at 16:31
  • Alignment would be a pain, and using more heads for more platters would get you more capacity and throughput for your investment in heads and related circuitry. – Chris Stratton Sep 24 '20 at 16:32
  • @BrianDrummond: Nice reference. Yes, Burroughs systems had HPT disk, and used it mainly as the backing storage for virtual memory because of its speed. Conventional moving-head storage was used for user files. Storage *density* sucked, however, because there's a limit to how many heads (tracks) you can pack in there. – Dave Tweed Sep 24 '20 at 18:09
  • @evildemonic Hmm - interesting, but I was referring to having multiple ARMS. Or is the performance added by having multiple arms negligible? – SPQR Sep 24 '20 at 18:14
  • You are asking about increasing performance. My new pc has a solid-state drive that produces immediate results. – Audioguru Sep 24 '20 at 18:28
  • @SpqrTiang That is addressed in the accepted answer on my link. Specifically the 3rd and 4th paragraphs. Basically, it's just not worth it, the small increase in performance is heavily outweighed by the added complexity and cost. There are easier ways to increase performance. – evildemonic Sep 24 '20 at 18:47

1 Answers1

2

It's been done, but not so much anymore. Early drives were physically large, and used a very simple servo platter to ensure tracking, so multiple actuators was very feasible. At the same time, these big drives had longer seek times and lower RPM, so latency was a problem and therefore worth the effort and expense of dual actuators.

With today's smaller drives using advanced high density track encoding techniques it would be very difficult (if not impossible) to get mechanical coherence between multiple actuators on the same platter. It also increases the complexity of the drive which impacts reliability. Finally, smaller drives have much shorter seek times than those old washtubs so there's less penalty when the head has to hop around. The juice (and added cost) just isn't worth the squeeze with modern drives.

So why did they do it at all way back in the day? The main reason is dual-porting the drive, with a secondary benefit of increasing I/O throughput. This is useful in larger systems with multiple hosts to support fail-over and increase computation throughput.

These same reasons are valid today, but done differently. The dual-port idea persists in SAS (Serial Attached SCSI) at the connector level, again, to support failover and shared working storage. Dual-port SAS drives have two logical interfaces, but nevertheless use only one actuator, with accesses coordinated and scheduled by the drive controller.

Finally, for very latency-sensitive applications, there's the option to use SSD drives, which are getting cheaper and more reliable all the time. These too can be dual-ported for scalability and failover support.

hacktastical
  • 49,832
  • 2
  • 47
  • 138
  • Seagate has recently started providing 2-actuator disk drives: https://www.seagate.com/innovation/multi-actuator-hard-drives/ (This effectively provides two logical drives within a single disk enclosure while sharing many hardware components.) –  Jun 28 '21 at 18:31
  • Looked at this. The actuators are on separate platters. – hacktastical Jun 28 '21 at 18:38
  • Using separate platters does reduce the rotation benefit, but compared even to single-package dual drive avoids the costs of separate spindles and provides better I/O per second for random reads (equivalent to adding more disks). For random reads doubling the number of drives seems to come close enough to doubling the number of heads per platter that the cost benefit dominates. I agree, it was not what the question was specifically addressing, but it does provide some of the advantage. –  Jun 29 '21 at 18:03