· Content
· News
· Articles
· Mailinglists
· Knowledgebase
· Trouble Tickets
· Files
· Glossary
· Links
· Compatibility Lists
· Forums
Welcome to our website
To take full advantage of all features you need to login or register. Registration is completely free and takes only a few seconds.
RAID-5 On The Desktop - Hardware vs. Onboard Compared
Posted by: duke on: 01/10/2005 03:17 PM [ Print | 6 comment(s) ]
GamePC has thrown together an interesting comparison of onboard, software-based RAID controllers and add-in hardware-based controllers. Which would better suit your needs and budget?
Onboard RAID-5 technologies help ease the amount of setup by the end-user (as RAID-5 arrays can be setup quickly in the BIOS, instead of in the OS) and are a bit more flexible. Onboard RAID-5 controllers still typically divert RAID-5 processing to the system's main CPU, which can lead to performance slowdowns in heavy situations. Still, considering these onboard RAID-5 solutions are
« Exploit code attacks unpatched IE bug · RAID-5 On The Desktop - Hardware vs. Onboard Compared
· Intel and AMD's 2005 Plans Revealed »
Comment
|
Contrablue Registered User Posts: 68 Joined: 2004-12-21 |
#33860 Posted on: 01/11/2005 12:58 AM
Greetings, I was surprised that GamePC didn't test the Areca SATA RAID controller -- after all, it posted spectacular scores in a recent review in tweakers.net (see http://www.tweakers.net/reviews/525/3). So I decided to use the same benchmarks GamePC used on my own Areca-based RAID5 setup to get a feel for how the GamePC results would have looked with the inclusion of the Areca. Before I enumerate the results of my tests, I should mention that my configuration differed a bit from the one used in GamePC. GamePC's configuration used a 3.2 ghz Pentium w/ 1gb SDRAM; they used a 10K RPM Raptor drive (the fastest SATA drive available at this writing) for the OS, with a separate RAID 5 array made up of 3x Maxtor DiamondMax 200gb drives. My own configuration (see sig below) differed in that I used a 4x drive array (this would give me a possible 20% transfer rate advantage), but I have no separate drive for the OS (which would somewhat offset my transfer rate advantage). Also, though I have dual processors, I'm using Opteron 244s which are thought to have performance equivalent to, at most, a 3.0 ghz Pentium IV. Another difference is that my OS is 32-bit Windows 2003 Server (I used the older SCSIPort driver), whereas GamePC used 32-bit Windows XP SP2. Given that I used SCSIPort, I don't think the OS made a significant difference in the scores. Here are my results (in parenthesis I included the GamePC's results for the best scoring RAID card in that test). Areca 1110 PCI-X SATA RAID Controller w/ 4x Seagate 7200.7 200gb drives in a RAID 5 Array DISK SPEED 32 Access Time: 12.5 ms (not tested by GamePC) Read Burst: 574.3 mbps (SI 3114 -- 106.9) Peak Transfer: 304K (Broadcom 4452 -- 92K) Avg Transfer: 220K (SI 3114 -- 83K) HD TACH Long Test Burst Xfer: 558.7 (SI 3114 -- 104.9) Avg Xfer: 151.3 (SI 3114 -- 91.9) CPU Use: 4% (3Ware 9500 -- 1%) Access Time: 12.5ms (3Ware 9500 -- 13.3ms) DISCUSSION My system, due to the addition of one more drive to the RAID 5 array, but with some offsetting disadvantages, should have had an advantage of perhaps 10% in these benchmark tests -- most of the other important configuration parameters between the GamePC test system and my rig are comparable. The big difference, it seems, between my system and the configurations tested by GamePC was in the SATA RAID controller -- which is what we're trying to test anyway. But the Areca controller did so much better than any of the controllers tested by GamePC -- 5 times faster burst transfer, 2 times faster average transfer -- that I began to wonder if there were other differences between my and GamePCs configurations. There is one significant configuration parameter that GamePC did not document in their benchmarks: the speed and width of the PCI bus to which they connected their tested controllers. In my case, I plugged the Areca in to an available 64-bit 133mhz PCI-X slot on my DK8X mobo. I'm pretty sure that the ASUS mobo used by GamePC has comparable PCI-X slots, but it's not clear that GamePC used one. If GamePC used a non-PCI-X slot, or used a 32-bit PCI slot, then they potentially hampered 3 of their 4 test results (the Silicon Image 3114 is built in to the ASUS mobo) -- possibly with very severe consequences. I'm assuming that the people at GamePC are more expert than me in testing, and would not have made the mistake of putting a high-performance RAID adapter like the 3Ware Escalade into a 32-bit PCI slot. If this is the case, the Areca SATA RAID controller deservers much more attention (and consideration by members of 2cpu especially) than it was given in the the probably flawed Tom's review or the glowing Tweaker's comparison. Regards, Contra DK8X, 2x Opteron 285, BFG 6800 Ultra, 8x 512mb Corsair Twinx PC3200, Areca 1110 SATA RAID 5, 4x Seagate 200gb 7200.7 |
Comment
|
FrostByte Registered User Posts: 39 Joined: 2002-08-14 |
#33861 Posted on: 01/11/2005 06:08 AM
Hello I think it's an Asus P5AD2 Premium Socket775 since this board has an onboard Silicon Image 3114 controller with software RAID5 support http://http://www.asus.com/products4.aspx?l1=3&l2=11&l3=0&model=73&modelmenu=1 thus the board has no PCI-X slots just a legacy PCI, a pair of PCI-express 1x and a PCI-express 16x slot I was under the impression GamePC was reviewing RAID5 as a desktop solution not intended for high end workstation/entry level server markets FrostByte "you can order a computer on a computer from a computer, and now your talking to a computer, show me the humans!" |
Comment
|
Contrablue Registered User Posts: 68 Joined: 2004-12-21 |
#33862 Posted on: 01/11/2005 08:48 AM
Hi Frostbyte - I went over to the ASUS site you linked and checked out the motherboard they used in their recent test of RAID controllers. You appear quite correct -- the motherboard used by GamePC in their tests almost certainly lacked 64-bit PCI OR PCI-X 133 connectors. In plugging a modern, high performance RAID controller (and the three non-mobo solutions all qualify) into a 32-bit PCI slot, the GamePC reviewer unwittingly invalidated his review by: (1) Strongly biasing the results in favor of the onboard SI 3114 (which in other benchmarks has generally performed far below the hardware RAID solutions) (2) Operating the hardware RAID cards in manner not recommended by their manufacturers. From the 3Ware 9500 Manual: "Although the controller fits in both 32-bit and 64-bit PCI or PCI-X slots with 5V as well as with 3.3V, install it in a 64- bit slot to take full advantage of the controller DK8X, 2x Opteron 285, BFG 6800 Ultra, 8x 512mb Corsair Twinx PC3200, Areca 1110 SATA RAID 5, 4x Seagate 200gb 7200.7 |
Comment
|
pwdrhnd23 Registered User Posts: 14 Joined: 2002-10-11 |
#33863 Posted on: 01/11/2005 10:14 PM
I would have liked to see the software based raid as well, just to see if these controllers merit their price tags. I have never seen any numbers comparing the software raid-5 to hardware, at the very least it would be a good standard to measure other implementations. |
Comment
|
NegativeEntropy Registered User Posts: 39 Joined: 2001-06-02 |
#33864 Posted on: 01/12/2005 09:40 AM
Um, is it just me, or should any RAID-5 performance comparison spend at least as much time measuring/evaluating write performance as read performance? Historically, write performance is the weakest point of RAID 5 and they had one (1!) synthetic benchmark "measuring" it (without benchmark settings even documented). Disappointing. |
Comment
|
shoarthing Duellist Posts: 3061 Joined: 2000-07-25 |
#33865 Posted on: 01/12/2005 12:45 PM
Asus P5WDH; watercooled E6420/PSU/GPU/[encapsulated]HDDs/NBridge/SBridge/VRMs; 4GB Elixir/Nanya PC6400; 2407WFP; mac OS X 10.4.10 |