Hello everybody!

Here is a way to build a hard disk twice as fast as now: build it with two arms!

The arms are at opposite locations with respect to the platters. Each platter track holds data blocks: #0 2 4 6... and 180° apart: #1 3 5 7... Both arms read or write the same track as soon as the data is over a block long.

So when the OS requests to read or write blocks #0 1 2 3, one arm reads #0 2 and the other #1 3, and since both the blocks and the arms are 180° apart, data is read simultaneously, thus doubling the contiguous read/write throughput. Get 240MB/s with current technology!

Even better, the platter's rotational latency is halved. That is, if reading blocks #0 2 by the North head and #1 3 by the South head requires to wait more than 180°, just swap the roles and read #0 2 by the South and #1 3 by the North. A 7200/min disk then has the same rotational latency as a 15000/min!

What does not improve with two arms is the time needed to move a head from one track to another. So with usual disks with a seek time of 8.5ms and a rotational latency of 4.2ms, random access time sinks from 8.5+4.2=12.7ms to 8.5+4.2/2=10.6ms, and this is the figure HdTach or HdTune will show - less impressive. Drop the innermost area of the platter to improve these benchmarks.

One should understand, however, that this time is very pessimistic because files needed by an OS or an applications aren't just scattered randomly over the disk. Since Windows 98, the files each application loads are observed by the defragmenter who then places them next to another on the disk.

Then, as the arm moves very little and hence quickly, disk rotation is the main latency of a disk. This is the performance measured by CrystalDiskMark for instance, and it will double with two arms.

XP and followers improve hard disk operation further by issuing parallel read commands for all files XP or an application need (keyword Prefetch), so that the disk can reoptimise the access sequence (keyword NCQ) depending on the files' positions on the platters. This is the performance measured by Atto for instance, and it doubles as well with two arms.

Optimizing multiple file accesses with two arms requires a special controller onboard the disk; two independent Sata connexions on a separate Raid-0 adapter wouldn't be quite as good. It's about time to issue the Sata-3 standard, we saturate Sata-2 here!

A two-armed disk has much better performance than a raid-0, because it halfs the average rotational latency instead of multiplying it by 4/3 - this is even better than spindle synchronization, and is compatible with adaptive formatting. Very small stripe sizes would make sense with Windows 2000 Pro and before, where parallel read commands are uncommon - another advantage over Raid. In servers and XP, bigger stripes make sense thanks to simultaneous commands.

Two arms need a longer disk case, and this is a huge drawback for 2.5" and 1.8" disks, where Compact Flash and SSD are likely to exterminate hard disks. Put a 1.8" platter in a 2.5" case? But for 3.5" disks, things look very good. Many PCs have room for a longer disk case, and any PC enthousiast would adapt a disk tray to double the disk performance. As for a file server, people build the computer around the disks anyway.

This means that a two-armed disk can reuse existing platters and existing arms and heads - the difficult technology in a disk. Only the controller and the case must be adapted. All OS and applications can remain the same. The improved performance should crush Flash speed; combined with terabyte capacity, this would ensure some years 3.5" hard disk supremacy over Flash competitors.

Of course, a 10,000 or 15,000 /min disk can be built with two arms as well. And more than two arms would improve performance further (organize blocks accordingly); this needs a faster interface, and a disk case broader as well - not a significant drawback for a file server.

I hope to be the first inventor of the multi-armed disk, or at least of the corresponding block organization that reduces the rotational latency. I have no patent on it and it is public from now on, so please feel free (but check existing patents before, I didn't) to use this idea to provide me with faster hard disks.

Marc Schaefer, aka Enthalpy

For those interested in disk benchmark, here are download addresses.

Hd Tach and Hd Tune measure a random access time over the whole disk:
http://www.simplisoftware.com/Downloads
http://www.hdtune.com

WinBench is less easy to use but is accurate on Raid-0 and measures access time within a partition:
http://www.benchmarkhq.ru/english.html?/be_common.html

Crystal Disk Mark measures random access within a smaller area like 50MB or 100MB:
http://crystalmark.info/software/CrystalDi...rk/index-e.html
without parallel commands.

Atto measures it within a smaller area ("Total length", choose it bigger than the buffer!):
http://www.hugesystems.com/supportspace/bench32.exe
and it issues several commands in parallel ("Queue depth") with Nt4-2k-Xp.

The last two are useful for Compact Flash as well, as they observe the write speed of small files - this explains some disappointments with operating systems on Compact Flash.

I've put a drawing of a two-arms version there:
http://saposjoint.net/Forum/viewtopic.php?f=66&t=1793
which should help understand the explanation.
Enjoy!

Marc Schaefer, aka Enthalpy

In addition to the drawing of the multi-arm disk there
http://saposjoint.net/Forum/viewtopic.php?f=66&t=1793

a way to manufacture lighter, faster arms there
http://saposjoint.net/Forum/viewtopic.php?f=66&t=2160

I just hope manufacturers of mechanical disks won't give up!
I still need fast disks!

Conner did this some time ago.
High cost for limited improvement ans so didn't really work out that well from a business perspective.
Of course, now there is also RAID striping which allows you to increase the number of heads processing your data using multiple drives and giving you disk failure redundancy to boot.

http://www.tomshardware.com/news/seagate-h...drive,8279.html

Arthur

A two-armed disk would be way better than a Raid array, because
- It cuts the average rotational latency in half, instead of increasing it by a factor of 1.5, when requesting a big file;
- It cuts the average rotational latency in half, instead of keeping it unchanged when requesting small files.

As throughput has improved a lot over the years, but latency not so, latency is the limiting factor on most uses of a mechanical disk. Which explains why a Raid (with current 7200rpm and 500GB/platter) is experimentally slower than a single disk when launching an OS or an application, and why a multi-armed disk would be faster.

Isn't this all rather irrelevant now that solid-state hard drives are becoming more prevalent?
It doesn't matter how many arms you throw on a hard disk platter, it will never match the potential of a solid-state drive.

Well done Adoucette! Conner called it Chinook, as stated here:
http://en.wikipedia.org/wiki/Conner_Peripherals

Looking for reasons why it should work better now...
- They probably put the arms at opposed locations. If not for performance, then for room. Same design hence.
- But did they put the stripes spread at 180°? That is, 0246... starting at 0° and 1357... starting at 180°? This is the trick for reduced latency.
- In 1995, drives had 3600rpm to 7200rpm, but their throughput was way less than half the current 134MB/s. So latency was less critical then.

I own an excellent SSD but I want to have fast mechanical disks as well, of course.