Hitachi Deskstar 7K1000 in RAID 0: Is Two Terabytes really better than One?
by Gary Key on April 19, 2007 12:15 AM EST- Posted in
- Storage
iPeak Video/Audio Tests
The iPeak based Video/Audio benchmarks are designed around simulating media encoding and HTPC activities. These are basic benchmarks at this time but this section will be expanded once we start testing under Vista. Our change to a better performing dual core processor will assist us in maintaining a balance between the CPU and Storage systems during the trace file creation and benchmarking processes. These benchmarks are CPU intensive in nature but also require a balanced storage system with the ability at times to handle read and write requests simultaneously in a very efficient manner.
The AnyDVD benchmark is heavily weighted to write requests. The results show a common pattern in write intensive situations with the RAID 0 setups improving about 66% over their single drive counterparts. The 7K1000 scores particularly well in these write intensive benchmarks due to its areal density and large cache design.
The Nero Recode 2 benchmark is weighted to streaming read requests at the start of the test but is balanced by continuous write operations during the encoding process. This benchmark is one of the most demanding ones in our test suite with the disk being active the entire trace file with several 100% utilization peaks. The results show an almost 93% improvement in hard disk performance with RAID 0. We believe this is one of the few tests where RAID 0 on the desktop will make an actual difference in real-world performance.
The individual drive results surprised us as we expected the Hitachi with its 32 MB cache and high sustained transfer rates to score near the Raptors. However, it turns out after reviewing the trace file results it was obvious that the 7K1000 was hampered by its inability to process large data blocks in sequential order efficiently as it had a number of buffer overruns during the read portion of the tests. This indicates to us the drive firmware is probably tuned for non-sequential read/writes as the rest of our tests indicate. The 32 MB cache should have easily compensated for any potential large block issues in this test.
iPeak Game Installation Tests
Our iPeak based Game Installation benchmarks simply show the ability of the hard drive to write data as quickly as possible to the disc based upon the installation software instructions. As detailed in our iPeak setup description we installed the games from our source drive in order to eliminate the optical drive bottleneck. In separate application timing we witnessed basically the same percentage spread when installing the games via our DVD drive so these results are representative of actual installation performance.
The Raptors once again finish at or near the top in our gaming tests due to their rotational and random access speeds. Our 7K1000 drive finishes ahead of the other 7200rpm drives with a great deal of benefit going to the 32 MB cache and high sustained transfer rates. In the RAID 0 results we see the Raptor scores improving 38% in The Sims2 and 55% in BF2. The 7K1000 benefits greatly from RAID 0 in these tests with improvements of 50% in The Sims2 and 90% in BF2 with the Raptors once again showing their strength in gaming.
We need to remember these tests reflect pure hard drive performance and will be mitigated by the overall system platform as we will see in our application tests. These tests are basically designed around continual read/write requests that favor large cache sizes, properly tuned firmware, and high sustained transfer rates.
iPeak Game Play Tests
The iPeak based Game Play tests are centered on the benefits of having a hard disk that can load non-linear or sequential data files quickly without interrupting the flow of the game.
In game play the RAID 0 setups hold a 7% to 12% advantage in these benchmarks. We noticed in the trace files that performance improved only when the game levels changed. The best overall gaming performance with a SATA drive is still provided by the Raptors. However, considering the space, noise levels, and cost per GB advantages of the Hitachi drive, we are willing to change our recommendation for most users (though plenty will feel 1TB is too big and would rather go with a 500GB model).
The iPeak based Video/Audio benchmarks are designed around simulating media encoding and HTPC activities. These are basic benchmarks at this time but this section will be expanded once we start testing under Vista. Our change to a better performing dual core processor will assist us in maintaining a balance between the CPU and Storage systems during the trace file creation and benchmarking processes. These benchmarks are CPU intensive in nature but also require a balanced storage system with the ability at times to handle read and write requests simultaneously in a very efficient manner.
The AnyDVD benchmark is heavily weighted to write requests. The results show a common pattern in write intensive situations with the RAID 0 setups improving about 66% over their single drive counterparts. The 7K1000 scores particularly well in these write intensive benchmarks due to its areal density and large cache design.
The Nero Recode 2 benchmark is weighted to streaming read requests at the start of the test but is balanced by continuous write operations during the encoding process. This benchmark is one of the most demanding ones in our test suite with the disk being active the entire trace file with several 100% utilization peaks. The results show an almost 93% improvement in hard disk performance with RAID 0. We believe this is one of the few tests where RAID 0 on the desktop will make an actual difference in real-world performance.
The individual drive results surprised us as we expected the Hitachi with its 32 MB cache and high sustained transfer rates to score near the Raptors. However, it turns out after reviewing the trace file results it was obvious that the 7K1000 was hampered by its inability to process large data blocks in sequential order efficiently as it had a number of buffer overruns during the read portion of the tests. This indicates to us the drive firmware is probably tuned for non-sequential read/writes as the rest of our tests indicate. The 32 MB cache should have easily compensated for any potential large block issues in this test.
iPeak Game Installation Tests
Our iPeak based Game Installation benchmarks simply show the ability of the hard drive to write data as quickly as possible to the disc based upon the installation software instructions. As detailed in our iPeak setup description we installed the games from our source drive in order to eliminate the optical drive bottleneck. In separate application timing we witnessed basically the same percentage spread when installing the games via our DVD drive so these results are representative of actual installation performance.
The Raptors once again finish at or near the top in our gaming tests due to their rotational and random access speeds. Our 7K1000 drive finishes ahead of the other 7200rpm drives with a great deal of benefit going to the 32 MB cache and high sustained transfer rates. In the RAID 0 results we see the Raptor scores improving 38% in The Sims2 and 55% in BF2. The 7K1000 benefits greatly from RAID 0 in these tests with improvements of 50% in The Sims2 and 90% in BF2 with the Raptors once again showing their strength in gaming.
We need to remember these tests reflect pure hard drive performance and will be mitigated by the overall system platform as we will see in our application tests. These tests are basically designed around continual read/write requests that favor large cache sizes, properly tuned firmware, and high sustained transfer rates.
iPeak Game Play Tests
The iPeak based Game Play tests are centered on the benefits of having a hard disk that can load non-linear or sequential data files quickly without interrupting the flow of the game.
In game play the RAID 0 setups hold a 7% to 12% advantage in these benchmarks. We noticed in the trace files that performance improved only when the game levels changed. The best overall gaming performance with a SATA drive is still provided by the Raptors. However, considering the space, noise levels, and cost per GB advantages of the Hitachi drive, we are willing to change our recommendation for most users (though plenty will feel 1TB is too big and would rather go with a 500GB model).
Facebook
Twitter
RSS
48 Comments
View All Comments
Gaelstorm - Wednesday, October 8, 2008 - link
This article was a long time ago, but the second to last comment before mine was talking about fair. If a program is not hard drive intensive and is just slow all around like the Sim 2, you are not being fair by mentioning that Raid 0 doesn't help it. Honestly nothing will help with that games load time except a faster computer overall. Even so that game loads horrible for the type of game it is. I can tell you right now, there are very few situations I have encountered where Raid 0 didn't way outperform a single drive for gaming or just feel on a desktop type system. All the arguing I read, actually made me rethink trusting information from certain sources at all.Per Hansson - Tuesday, April 24, 2007 - link
Some ideas, as could be seen on page 4 of the review the system was limited by something (flat graph in STR instead of what you would expect)This is probably because you are using the normal SCSI Miniport driver, which happens to be the only choice for Win2K and WinXP 32 bit
WinXP 64bit, 2003 Server 32bit and 64 bit and Vista 32 bit and 64 bit has the new StorPort miniport driver, which improves performance tremendeously in RAID arrays
Of course the controller cards driver must also actually support and use the StorPort driver instead of the normal SCSI miniport driver
What this means in laymens terms is that it was not because the "software based" RAID controllers where crap that we got shitty performance in Win2K and XP, it was because those controllers had shitty drivers. Which really was MS fault.
The really expensive RAID controllers however did not use the NT4/2000 Server etc native SCSI miniport driver but had their very own implementation (just as good as MS "all new" storport driver) which meant that they where not capped at about 100-150MB/sec, but rather unlimited... Try your review with 4 150GB Raptors in RAID-0, your performance will be the same because it is the drivers that is holding the system back...
Please keep this in mind for your upcoming review ;)
http://download.microsoft.com/download/5/6/6/5664b...">http://download.microsoft.com/download/...06-45ec-...
ShadowdogKGB - Saturday, April 21, 2007 - link
You have 5 synthetics that tell us that two are faster in stripe but then you have 5 real world tests that aren't real world at all. Tell me who plays HL2 Lost Coast and Sims2 Neighborhood. I think you should just stop testing raid setups all together since you can't be honest about it.Gary Key - Sunday, April 22, 2007 - link
The Sims2 continues to be one of the top selling/played games in the world and HL2 along with its various versions is also a top played title. We are revising our benchmark suite to include WoW and a couple of other titles that were recently released. As far as the other benchmarks, compression/decompression and encoding are fairly common activities on a PC the last time I checked. ;-) What are we not being honest about?
jleboeuf - Saturday, April 21, 2007 - link
I'd be interested to see what your results fair out w/ a raid 10 config w/ 6 drives. that's 3 terabytes, and completely mirroredaethyrmaster - Friday, April 20, 2007 - link
They said they only used a 64 kB (KiB if you prefer) stripe size. Using stripes of 128 kB size would most likely have changed the results significantly. If possible, I'd like to see results with a 256 kB stripe as well.Separately, I personally have used RAID 0, as well as used exactly the same drives in a non-RAID setup. I often times do a lot of transferring of video from my PC to other devices on a Gigabit ethernet network, and the extra 10 to 30 MB/sec transfer rate provided by RAID 0 is well appreciated when you are copying 400 to 800 MB (MiB) of video.
GOSHARKS - Friday, April 20, 2007 - link
Is that even statiscally significant? As it is, it can be slightly misleading.
Gary Key - Sunday, April 22, 2007 - link
Not in the grand scheme of things but the tests were run five different times with the same results. We were providing information based upon having a retail drive to test instead of the OEM drive we had in the first article. There were a few comments that suggested differences could have existed between the two so we tried to answer it.
tshen83 - Thursday, April 19, 2007 - link
reviewing a 1TB drive by raid 0? the writer obviously have NO FREAKING IDEA of what the target audience is. A RAID5 review would be much more appropriate. Hate to lose 100GB of data nowadays...let alone 2TB, at that point, you might as well kill yourselfGary Key - Thursday, April 19, 2007 - link
We have a full RAID article in the works that will test RAID 1, RAID 1+0, and RAID 5 with several drives, Intel and NV chipsets, along with hardware controllers. As I mentioned in the article, this story will be up in the coming weeks. As far as offering RAID 0 results with a motherboard controller chipset, this was done as the vast majority of boutique systems offer this option along with several hundred emails asking when we could provide RAID 0 results with the latest drive releases.