Which is quicker – A RAID Array or a Raptor?

March 18th, 2008

With desktop computers at the moment non-volatile storage options are fairly limited – and for everyday use the humble hard disk drive (HDD) is still the weapon of choice.

Hard Drives are proven performers – they’ve been around since 1956 – but they have a few limitations, mostly related to the fact that they have mechanical components:-

1. They fail – eventually, and usually catastrophically.
2. Their access speed is limited by the speed it takes the head to physically move across the platter and find the data you’re after – called ‘seek time’

One approach to overcome these limitations is called ‘RAID’ – or a Redundant Array of Inexpensive Disks (some folks argue that Inexpensive should be Independent, but I’ll beg to differ).

In this post, I’ll be investigating ways to speed up your computer – and I’ll be test driving RAID and the fastest available single drive solution (Western Digital Raptor) along the way – let’s fish out the facts from the hype.

What is RAID?

There are several flavors of RAID, but the major ones are RAID 0, RAID 1 and RAID 10.

RAID 0 addresses the second limitation – disk access speed – basically it spreads your data across various disks (usually two but can be more) in what is known as ‘stripes‘ – basically chunks of a file. These chunks are usually quite small (64K is typical). Basically what this does is theoretically at least double the speed to access a file – you have at least two hard drives sharing the load, and as we know, two heads are better than one J

RAID 1 addresses the first limitation – failure. RAID 1 ‘mirrors’ data. Basically, when you save a file on your hard drive, RAID 1 seamlessly saves a copy on a second hard drive behind the scenes. If one drive happens to croak, the second should still have a copy, so you don’t lose your data. It’s essentially an automatic backup, and adds redundancy to your system.

RAID 10 combines the best of both worlds – it both stripes and mirrors your data – so you have a backup as well as speed boost – but you need a minimum of four disks to implement a RAID 10 array, as opposed to two for the others.

RAID has been used for ages in servers – only recently with the advent of quite quick RAID chipsets (such as the intel ICH9R) on consumer grade motherboards has it really become an option for the home user.

Alternatives to RAID for fast disk access

One other thing that you can do to speed up disk access is to increase the speed of rotation of the hard drive. The standard for a 3.5″ disk these days is around 7200 RPM – but you can get 10,000 RPM drives from western digital, and they’re known by the trade-name ‘Raptor’

The Raptor’s offer around a 25% read speed increase over 7200RPM drives for non-fragmented data on account of the fact that they rotate faster.

So which is better? RAID or Raptor?

Well – that’s a difficult question, but I can talk about the theory. In theory, A RAID 0 or RAID 10 array should be better than a Raptor because they double (or more) the bandwidth from the drives – so essentially a 2 disk RAID array should (under perfect circumstances) perform with similar speed to what a 14,400 (2*7200RPM) drive would – a raw theoretical speed increase compared to the Raptor of 50%.

Of course this relies upon a number of suppositions:-

1. That the disks are the same size and density (the speed that the media rotates under the hard drive head decreases from the periphery of the hard drive as you reach the center – this is because of the principle of ‘tip speed’ – basically imagine you have a truck tire and a mini-moke tire next to each other. You roll each along the ground 1 revolution – which goes further? The truck tire of course – it has a larger circumference – the truck tire goes further per revolution than the mini tire – so, for a given RPM, the truck tire goes much further than the mini – it has a higher tip speed.
2. That the actual speed of the head is equal between both disks.
3. That there are no serious ‘bottlenecks’ in the RAID controller that limit data transfer speeds.

So let’s test this supposition.

Trial Setup

For the purposes of this trial, I took a normal desktop PC with a Gigabyte EP35C-DS3R motherboard (with intel ICH9R RAID Controller on board) with 2GB of RAM and a Q6600 Quad Core processor.

I first added a standard 7200RPM hard drive, and loaded it with windows XP – basically I didn’t want the hard drives slowed down by the OS at all, so I put it on a separate drive altogether.

I then consecutively tested each of the following as a ‘secondary’ drive:-

1. 1 x 10K RPM 150G SATA Hard Drive
2. 4 x 7.2K RPM 500G SATAII Samsung HD501LJ hard drives in RAID 10 (64kb stripe)
3. 2 x 7.2K RPM 500G SATAII Western Digital WD5000KS hard drives in RAID 0 (64kb stripe).

I then used a free utility called ‘HD Tune‘ to test the read speeds for each arrangement.

Results

In these results I’ll be talking about two things –

1. Access time (the average time it takes for the head to move across the platter and find the data we’re after)
2. Transfer Rate (the speed data is transferred in MB per second from the drive to the computer)

On the following graphs the access time is represented as a series of yellow dots, with the time on the right hand y axis.

The transfer rate is represented as a blue line, with values on the left-hand y axis.

The x axis shows the position of the head on the disk – 0% is the outside of the disk, 100% is the inside of the disk (remember, tip speed, and hence data transfer rate, decrease as we near the centre of the disk)

Test 1 – 10,000 RPM 150G Raptor.

Test 2 – 4 x 7200 RPM Samsung SpinPoint 500GB hard drives in RAID 10

Test 3 – 2 x 7200 RPM Western Digital 500GB hard drives in RAID 0

Results

In a Nutshell –

Access Time

The Raptor had a lower average access time across the platter than either of the RAID arrays – a good thing, BUT – there is a density difference between these drives.

If you consider that the RAID arrays are 1TB in size whereas the Raptor is 150GB in size, and look at the graphs for the RAID array you’ll see that the first 150GB of these drives (0 to 15% on the x axis) have very similar access times (probably due to the fact that the data density is higher in these drives) if not identical to the raptor.

This tells me that the heads are not as speedy on the WD and Samsung as on the raptor, but the greater data density on the high-tip speed periphery of these drives makes up for it.

Transfer Rate

The Raid Arrays comprehensively kicked the Raptors backside in sheer transfer speed – The average transfer speed was between 50 and 65% higher for a 2 disk 7200 RPM RAID Array than a Raptor Across the drive – which is in keeping with our theoretical calculations.

Additional Results –

• There was no noticeable decrease in read performance between the RAID 10 and RAID 0 Arrays.
  

Recommendations – Should I use a Raptor or a RAID array?

The lower seek times for the Raptor could be useful for quick access to fragmented data – the operating system for example. On the other hand, the seek speed of the RAID arrays were just as good as the Raptor around the first 150GB of the disk, with far superior transfer rates.

If you take into account the massive cost of the Raptor (a 150GB Raptor is around the same price as two 500GB Samsung Hard Drives) it’s almost a no-brainer – go the RAID – for the same coin you’ll have 4.5 times the storage space and a much faster system – provided your files are not tiny or highly fragmented.

If you’re worried about the reduced seek speeds of the RAID array, consider partitioning your drive and using the outside sectors of the disk to host your OS.

In my case, I have two setups now – both with the same processor.

In the first, I’m using a Raptor to host my OS and a RAID 0 Array to host my documents. In the second I’ve got an unpartitioned RAID 10 array with everything – operating system and files on the one RAID volume.

I’d have to say – the RAID 10 only volume boots slightly quicker and feels generally ‘snappier’ than the Raptor based system.

What I’d really like to test is a Raid of raptors J

All the best,

Matt

Entry Filed under: Technical Tips