Nearly 18 months ago, we published a review (link) of the then new Seagate 7200.11 1Tb, comparing it with the previous generations. Since then Seagate has stayed in the headlines for the 7200.11 series – and not for the best of reasons. Defective firmware leading to inaccessible drives caused a major splash and anecdotal evidence of large numbers of drives being RMA’d have steered popular opinion away from Seagate.

Many have criticized Seagate’s handling of the firmware problems, with initial slowness to publically address issues, fuzziness on which drives were affected and culminating in the vendor offering free data recovery services for those with bricked drives, with advanced exchanges for select customers. Our own Rage3D Administrator Lupine chronicled his issues with affected drives, RMA’ing problem drives and receiving drives with the same problem firmware back.

In the meantime, I had left my four Seagate 7200.11 1Tb drives doing backup duty as a RAID 5 set in an external enclosure. The original SD01 firmware stood the test of time, with no random rebuilds, dropped drives or poor performance seen – workhorses, all around. Moving to a new backup architecture, the drives became free once more for a revisit: new firmware, SD1A. Like before, no extravagant testing methods were used – just throw HDTach’s long bench at it three times and average the results.

I connected the four ST31000340AS drives to my Gigabyte GA-EP35-DS3R’s Intel ICH9R controller, and used Vista Ultimate SP1 x64 to run HDTach for some simple numbers.  Like the previously used discrete LSI and Adaptec controllers, Intel’s ICH matrix integrated raid solution offers a Write Cache. Initially this mode was introduced for RAID 5, where writes are coalesced to reduce the number of I/O’s per write for parity calculations for sequential data streams. This means if four sequential 16Kb access requests have been made, the 16K I/O’s are combined into a single 64Kb access. Intel Matrix Storage utilizes a cache for write buffering and to improve coalescing for desktop usage models, such as Raid 0 or 1, and AHCI. Approximately 4Mb of main memory is allocated for cache at boot time. Recent driver updates have allowed this write cache to benefit AHCI as well as RAID volumes, such that all IMS arrays benefit for burst read rates – in benchmarks, at least. As the writes are cached, subsequent reads from that same data will come from cache and not the disk, increasing data access speed. Obviously, this cache is sensitive to system configuration – the faster and lower latency the system memory, the better the cache performs.

Ideally the cache should be write-though – the data is written to the disk and kept in the cache for faster reads, vs. write back mode where data is flushed from cache periodically. In this mode data is very sensitive to power reliability and system stability; a bad overclock, system crash or a power interruption could not only cause data loss but operating system corruption if critical files are not written to disk.

To account for this, the tests were performed with the write cache enabled, and disabled, via the Intel Matrix Storage Manager. The CPU used in the system was an Intel Xeon X3210 B3 quad core processor, clocked at 3Ghz, with 4Gb of Buffalo Firestix DDR2 ram running at 900Mhz 5-5-5-15.

Seagate claims 105MB/sec maximum transfer rate for the ST31000340AS, which was verified in HDtach for the first 20% (200 GB) of the drive, in single disk configuration. Interestingly, the ST31000333AS 1Tb and ST3150041AS 1.5Tb drives are rated at 115MB/s and 120Mb/s, respectively. Those of you taking advantage of the recent sales on 1Tb and 1.5Tb drives are getting some fast drives, cheap.

Single and Two Disk configurations:

Here are the results for single disk and RAID 0 from the previous review Vista x64 benchmarks and the new flashed firmware, with write cache enabled and disabled:

No significant difference seen, certainly the differences in burst or average speed can be accounted for in the variations of driver and controller used. The single disk SD1A firmware appears marginally faster that the SD01.

Looking at RAID 1, again no big surprises:

Here I again included the single disk results to allow a simple contrast of RAID 1 with single disk. How much do you lose with the added redundancy? Hard to say as this is not a write test, but interesting to see how much benefit the RAID 1 performance gains from Intel Write Cache. Indeed, the Intel RAID 1 mode uses both disks for simultaneous reads, allowing different data to be read from both disks at the same time.

Three Disk Configurations:

These results compare the previous review’s Vista x64 results for RAID 0 and RAID 5 with updated firmware disks in Raid 0 and Raid 5. The previous review highlighted an issue both the LSI and Adaptec controllers had under Vista x64 with large parity configurations.

Using the ISM the SD1A drivers are clearly performing better. Again the use of write cache is noticeable, not only in burst results but in sustained average reads as well. CPU usage is low, 3-4% without write cache enabled and 6-10% with write cache enabled.

Four Disk Configurations:

Finally, we contrast the four disk results. These were the most problematic in the previous review – Vista 64bit driver issues, >2Gb volume issues under 32bit Windows XP. With the Intel ICH it was a breeze to generate results. Throughout the tests, the Intel Matrix Storage volume was initialized as a basic disk using GPT and a single volume created and quick formatted before bench testing.

Here, with write cache disabled, CPU usage stayed low at 3-4%. Enabling the write cache saw a jump to 11% for RAID 5 and 13% for RAID 0, with RAID 10 at 8% cpu usage. It would be interesting to see how CPU usage with write cache scaled up with five and six disk arrays. The increase is CPU usage makes me wonder if more than 4MB cache is used, depending on the size of the array and number of members. In terms of speeds the SD1A firmware drives have a strong showing here, mostly due to mature drivers I think.

The firmware SD1A certainly doesn’t appear to have harmed the ST31000340AS drives. Performance is robust, and reliability unaffected thus far. Current pricing makes them a very attractive buy, but certainly not the only choice – currently Samsungs 1Tb drive is retailing under $85 USD, and Seagate’s own new 7200.12 series 1Tb is $99. I’m still happy with my 1Tb 7200.11 drives, but I won’t be buying any more. I’ll be looking closely at the WD Black series and the 7200.12, mainly because of warranty, but also performance. Seagate recently changed their warranty terms away from the blanket 5-year for all OEM, Retail and Enterprise drives, excluding some to be 3 year. Until digital distribution and storage of media is widespread and commonplace, 500Gb drives and larger are going to be useful and likely in service for a long time. With that in mind, I’d certainly expect to be using a drive I bought recently, at 1Tb or greater capacity, in four or five year’s time as they roll down from my storage server to my desktop machines. For performance, the new 7200.12 series use a 500Gb platter, that is quieter and faster than before. Claims of 130MB/s maximum read performance are seen in the literature, with lower power, quieter performance and smaller form factor (for the single platter drives).

So how am I using these four ST321000430AS drives? In my HTPC/NAS with three disks in Intel Matrix Storage RAID 5 with the fourth disk as hot-spare and auto-rebuild target, running on a UPS. While no RAID array replaces a backup solution, it can give immense peace of mind for storage security. Load times for UT3 aren’t bad, either…