With the recent release of the Barracuda 7200.11 1Tb drive, Seagate claims 105 Mb/sec. sustained data rates. I thought it would be fun to compare a couple of the previous generations’ performance characteristics, in some simple benchmark testing.

At my disposal I had three sets of four Seagate Barracuda drives. Representing the 7200.9 era is the ST3250624AS, a 7200rpm 250Gb drive with 16Mb cache and a claimed max sustained transfer rate of 76.6 MB/sec. From the 7200.10 stable we have the ST3500630AS, again 7200rpm. Capacity is increased over the 7200.9 to 500Gb, but cache remains the same at 16Mb. Claimed max sustained transfer rate is 72Mb/sec. Last from the Seagate stable, I have the ST31000340AS representing the 7200.11 series. Again 7200rpm, but cache increased to 32Mb and a claimed max sustained transfer of 105Mb/sec. Finally, as a wild card, I have a WDFS version of the Western Digital 80Gb Raptor. This OEM only variant updates the venerable 1.5Gbps interface to SATA-II 3.0Gbps standards. Will the increased bandwidth make it even more attractive?

Testing was performed with HDTach 3.0.4.0, in both Windows XP SP2 and Vista Enterprise x64. LSI’s popular megaRaid controller was used, in the form of 8204ELP. This four port controller uses softRaid to provide Raid 0/1/5/10 levels, with compatibility for SATA and SAS drives in a PCIe x4 package. The host machine was a Dell Precision 690 equipped with 4x1Gb 667Mhz FB-DIMM’s and a Xeon 5160 (3.0Ghz 1333FSB Woodcrest) CPU. Some additional benching was done on two more controllers, the onboard Dell SAS5i/R controller (a re-flashed cut-down LSI controller itself), and an Adaptec 1430SA PCIe controller. The Adaptec and SAS 5i/R controllers were used to highlight controller performance vs. disk performance.

Each test was performed three times, and the best result used for comparison. It should be noted that 32-bit Windows XP had some trouble with arrays larger than 2Tb being reported correctly, leading to some slightly skewed results when taken in isolation.

Let’s begin with some single disk results, using the LSI MegaRaid controller for all tests:

We see very similar numbers for the drives, independent of OS version. The raptor really liked the increase in interface bandwidth, scoring the highest burst result. However, it plays second fiddle to the 7200.11, where the raptor’s combined average read (76.75 Mb/sec) trails the new Barracuda (87.2 Mb/sec). Both the 7200.9 and 7200.10 are trailing significantly in average read, but competitive with the 7200.11 for burst speeds.

For comparison, here is an IDE 7200.10 250Gb drive benchmark:

The SATA interface brings benefits for burst speeds, NCQ, hot plug ability and easier cabling & configuration. No difference for average read, but burst speed benefits tremendously.

Let’s look at random access times (where lower is better), for the complete picture:

Again, similar results for the Barracuda’s, with a expected lead for the 10,000 rpm Raptor.

So, what can we say about these drives in isolation? Well, for the best load times you want the WD Raptor. Duh. But as your home machine main drive? Much harder decision, as cost and space become important considerations, and 80Gb disappears fast with a Vista install and a few games, movies and downloads. Probably the best bang for the buck drive here is the 7200.10 500Gb drive, plenty of room for all your applications and data, and still nice performance. For a single OS drive the 7200.11 is expensive enough that you might consider two or more smaller drives in Raid, if you aren’t concerned about power consumption. So let’s start looking at Raid performance!

Raid 0 is an easy way to get more performance and consolidate storage from your drives. There is an increased risk of data loss, as if one drive dies all data on the other disk is unreadable. But if you’ve got no information stored on the array that you care about losing, or re-installing windows doesn’t cause you to cower in fear, it can bring some nice numbers to the table. All arrays in this test were using 64Kb stripe sizes.

Woohoo! All RAID sets scaled nicely here. The single disk combined average read scales up well on all pairs, but the new ‘cudas have an outstanding edge. Interestingly, the 7200.10 series is the new burst benchmark champion, but all the Barracudas show the strength of that 3gbps interface now.

How do random access times fare? Well, they should be the same:

Unremarkably, the access times are the same, with the Barracuda’s times roughly equal, and the Raptor’s extra 2800rpm gaining you approximately 4ms. Is the trade for read speeds worth the increased access time? Probably, especially considering you get a 2Tb array using 7200.11. You won’t be buying more for a while, especially not at the launch prices of the 7200.11 series. But what if you did? Before I answer that, I’m going to examine mirrored performance.

RAID 1 gives the first level of redundancy to a storage solution. Each disk is a copy of the other, giving you the ability to lose a drive and not lose any data, and schedule a maintenance window to replace the failed drive (or even run without it while you wait for RMA). The downside is the doubling of writes can increase latency in writes, and possibly reads under array failure conditions. Let’s look:

Nicely similar to the single disk benchmarks, meaning if the extra cost doesn’t matter, the redundancy is free. An interesting burst bench for the raptors under Vista x64, but it was consistent. At this point, I’m going to quit with Random Access times. They don’t change any, so lets just take it as read – 10krpm is better for random access. Until you start to think about 8ms for randomly accessing 80gb vs ~13ms for randomly accessing 1Tb…

Here’s where we lose the Raptors, as I unfortunately only had two to play with. But we can still have fun without those two hot, fast party animals, right? Back into non-redundant arrays here, 3 disks should give another boost in performance.

Now we’re talking! The proven 7200.10’s are starting to shine now, though the 7200.9’s are no slouch, either. Interesting the drop in performance under 64bit Vista now, quite significant for the 7200.11. Is this a driver issue for the LSI megaRaid? Looking at the Adaptec 1430 results (below) we see the same average read speeds, but a ~30mb/sec increase in burst speeds for the 7200.11; and a drop 50mb/sec burst for the 7200.10. As this was consistant across tests (all were repeated 3 times), I’m going to consider this a driver inconsistency for the controllers with 1Tb drives.

A number of onboard controllers offer this RAID level now, which claims the performance benefits of a striped disk configuration with the redundant protection of a parity check, stored across all disks. In high-end enterprise and server applications, the parity calculations are done by a dedicated processor with associated battery-backed cache to improve write performance and reliability. In onboard and low-end discrete cards (such as the LSI megaRaid 8204) these calculations are done by the host CPU. This could cause performance degradation if there are significant write transactions queued, however given the power of today’s modern CPU’s I doubt anyone would notice the 4% utilization I saw in my tests. Capacity is approximately 2/3 of the sum of total drive space, giving 500Gb, 1Tb and 2Tb arrays, respectively.

The net result is an array with improved write performance over a single disk or mirrored array, and substantially increased read performance, similar to a Raid 0 array. This is the basis of the ‘best of both worlds’ recommendations seen on many hardware enthusiasts forums. For interests sake I took a screenshot of a 3 disk hardware Raid5 from a Dell Poweredge 2950, with three 15krpm 76Gb SAS drives:

Those numbers don’t look that different, do they? Well, apart from the 5.8ms random access time! Of course, the HDTach benchmark doesn’t show how well the hardware Raid5 with cache & battery backup scales for large IO loads, but I think we can take it as read – for a home workstation/server, a dedicated Raid controller may be the point of diminishing returns. However, where the cache and dedicated processor will shine is under rebuild conditions, where the array loses its fault tolerant status – through disk hardware failure, power interruption causing an incomplete write, etc. When the array is rebuilding, the software Raid5 array’s performance will be significantly degraded, with read/write speeds ranging from between 5mb/sec to 30mb/sec. Not fun if you’re gaming. If you decide to use a Raid 5 array on your machine, make sure you monitor the array health, and let the operation complete before running disk IO intensive tasks, for best performance. Note that a full rebuild on a 3Tb array may take a significant length of time, depending on your processor. On the test system, a Verify of the 3 disk 7200.11 Raid5 configuration took 18 hours. Changing rebuild priority from low to normal decreased time to 6 hours, but disk performance was closer to 5mb/sec than 30mb/sec.

Back to basics – let’s stack up four drives, stripe ‘em and see how fast we can go! Here I noticed that large arrays (greater than 2Tb) weren’t handled very well by XP 32-bit. The raw array would sometimes be detected as the correct size, but other times would be limited to 2199Gb. Creating and formatting a GPT disk (Globally Unique Identifier Partition Tables are required for partitions larger than 2Tb under Windows) didn’t solve the issue. Even if the array was detected correctly, then HDTach wouldn’t report the correct size. Given that HDTach and Vista x64 didn’t experience these issues, the XP x32 results should be taken as unrepresentative of the 7200.11 drive as a whole, but show what the drive’s performance would be if partitioned to 2199Gb.

Again the 7200.9 and 7200.10 scale nicely, with average reads slightly higher and burst well in excess of the two disk stripe performance from the Raptor’s. This could be an interesting decision – four cheap widely available drives in Raid0 give more performance and space but slower seeks and higher power consumption, than two hot, noisy, expensive, small capacity drives. But four drives give you another Raid level option, which we’ll come to later. The four terabyte 7200.11 array looks to be underperforming, until you consider that number again – that’s four terabytes. The XP x32 benchmark shows that the first 2199Gb perform faster than the whole 7200.10 array, in both average read and burst performance. While the 158Mb/sec. average read rate of the 7200.9 array is nearly three times the single disk performance, so is the 7200.11’s – and we started so much higher. The 7200.11’s 210mb/sec average read is not far short of the single disk burst performance of all three drives. Pretty monstrous, if you ask me!

Moving on to Raid5 for some more parity fun. With four disks the capacities are approximately 750Gb for the 7200.9 array, 1.5Tb for the 7200.10 array, and 3Tb for the 7200.11 array. Windows XP x32 again reported the 7200.11 array as 2199Gb in Disk Manager, and through HDTach. Vista x64 correctly showed the 2997Gb size.

We see the effects of the parity calculation here, with all arrays average read speed decreased. The least performance hit is seen on the 7200.10 array, which loses a mere 10-12mb/sec from the Raid0 configuration. Is this a byproduct of the AAE firmware? Probably, as the comparatively high average read was evident with three disk Raid5 as well.

All arrays exceed two disk Raid0 and three disk Raid5 numbers, showing the performance does scale with more disks in Raid5. The misdetection of the 7200.11 3Tb array doesn’t seem to have affected the results of the tests, as the Vista x64 and XP x32 numbers are similar.

With four disks, a new Raid level is possible. Raid10 can be one of two subtypes. In Raid-0+1 data is organized as stripes across multiple disks, and then the striped disk sets are mirrored. In Raid-1+0, the data is mirrored and the mirrors are striped. The cost is half the disk capacity, giving 500Gb, 1Tb and 2Tb arrays respectively for the 7200.9, 7200.10 and 7200.11 disks. Windows XP x32 had no problems detecting these arrays correctly. This level of Raid is redundant and offers the best compromise of speed and redundancy, at the cost of available storage space. It is my preferred method of Raid array.

Interestingly, the 7200.11 starts to shine again here, with a significant lead in average read speed over the other two arrays. Burst speed is less than the 7200.10, but with capacity doubled I find the number very pleasing. However, my testing showed that Raid10 is not the LSI MegaRaid’s strongest point. A quick look at the Adaptec 1430SA shows that the driver & hardware does indeed make a difference.

All arrays exceed two disk Raid0 and three disk Raid5 numbers, showing the performance does scale with more disks in Raid5. The misdetection of the 7200.11 3Tb array doesn’t seem to have affected the results of the tests, as the Vista x64 and XP x32 numbers are similar.Phenomenal burst performance from the 7200.9 and 7200.10 series here, with respectable average read numbers, too. The low burst speed on the 7200.11 leads me to believe that the Adaptec 1430SA has issues with 1Tb drives, as the same low burst is seen under Vista x64.

Seagate has been my storage manufacturer of choice for a while now thanks mainly to their long warranty. Five years is a long time for a hard drive, and I believe it is a strong testimonial to how much Seagate believes in their product. The free tools available from Seagate for customers also are a nice added-value from the SeaTools product to check drive health and diagnose issues to the DiscWizard tool for adding your new drive and moving your data or whole install. DiscWizard is powered by Acronis and gives the power of a combination of both True Image and Disk Director products with excellent instructions.

If you have the power supply, fans and chassis space to implement it then arrays of cheap disks can give you high performance levels for relatively cheap. Whether you need the performance is a personal call, but bear in mind that the slowest part of modern personal computers is the storage. 3 GHz processors with 1333Mhz RAM and multi-GPU setups rely on getting data as fast as possible.

There are many factors to be considered when deciding on a storage solution. The first is to remember that no Raid array replaces a backup solution where your critical important data is stored in a non-volatile manner and is tested regularly. The purpose of a fault tolerant Raid array is to provide continuing service despite hardware failure and the opportunity to backup the most important recently updated data.

Secondly - power and cooling. More drives need more power and generate more heat. On power up all the drives will spin up loading up your power supply. At 20w average peak load per drive four drives take a not inconsiderable 80w extra draw at start up. All that power generates heat which means more cooling is required.

A Raid 0 array of two raptors has long been pimped as the best performance configuration for disk intensive applications. However as SATA-II technology developed the increased burst speeds made that decision tougher – especially as the loud, hot raptors are low on capacity by today’s standards. The OEM only SATA-II interface Raptors rectify part of the issue but the availability and still small capacity are a big detractor for the home builder. Sure, you could buy two raptors and a large storage drive, but price that against a three disk medium capacity striped array and the performance increase. Do you go with higher average read and burst speeds or faster access times?

Price per Gb is an important factor and both the Raptors and 7200.11 suffer here, though with the 7200.11 series this ratio will drop as supply increases. If you have the money and need the size they appear well worth it. However for smaller arrays a smaller capacity disk might be in order. With the prices of 500Gb 7200.11 dropping quickly I would have no hesitation recommending one over the 7200.10 series – as long as the 32mb cache was included.

So what did I do? I kept the two raptors in Raid0 as my main OS array for running Vista x64. I used the 7200.10 500Gb drives in Raid 10 on the Adaptec 1430SA as my data array - for storage of Acronis True Image backups of my OS and running virtual machines. Power and cooling aren’t an issue for the Precision 690 with a 1Kw PSU and a full tower size. The 7200.11’s went into an external eSATA enclosure in a Raid5 configuration connected to a server to store my server backups. The 7200.9’s are being distributed wherever they may fall to be used in individual machines.

The point that sticks with me the most however is how little performance hit a Raid1 array showed. For peace of mind, budget considerations, and simplicity, it is a solid choice.