EggXpert

The official Newegg tech support community and Newegg tech support forums. Learn about PC building, case mods, computer repairs, and computer troubleshooting. Get help from knowledgable community members about computer hardware and computer software, laptops, notebooks, netbooks, consumer electronics & mp3 players, home networking, lcd TVs, home audio and more.
Welcome to eggXpert.com. Sign in | Join | Help
in Search
Advanced Search

Sidicas

SB700 vs. 3Ware

These are benchmark results of a pair of 300GB WD VelociRaptors running on an SB700 RAID-0 setup vs. a RAID-5 of four 7200 RPM hard drives on a 3ware 9650SE RAID card.

 

Here is the benchmark from the 300GB WD VelociRaptors running in RAID-0 with 128KB stripe on an SB700 onboard chipset ( ASUS M3A78-CM ). I would have liked to use a larger stripe to improve random access performance, but these cheap onboard chipsets don't give you much options.

 

You see that squiggly line? That is a problem..  Really, the I/O should not be performing like that at all.  As the head gets closer to the center of the drive, the performance of the drive should drop accordingly, I believe the irregular squiggly line indicates that I'm having issues with the SB700 chipset / controller keeping up with these drives in RAID-0.  I'm running 64-bit Windows Vista with latest drivers, latest BIOS, latest patches, and I've tried it with native command queuing on and off.. I get the same results everytime..  The image above clearly shows that there is a problem in the I/O system that is not related to the hard drives themselves.   Also note the CPU usage in the graph, a little high at 12.5%, this is an AMD 5050e Dual-Core 2.6 Ghz.  Vista seems to do a great job at balancing the CPU across both cores and so this 12.5% is in fact a 12.5% load on two cores simultaneously.  Despite the CPU load, there is still something either in the I/O controller, in the drivers, or in the chipset that is preventing these drives from running to maximum capability..  Theoretically, each of these drives should be hitting 160MB/sec which means I should be at least 320MB/sec... (Edit: 10/8/09 See Correction below) If you take the tail end of the curve and arch the front part of the curve upwards to account for the performance difference due to the radial data storage location of the data, you can see where it "should" be.  It's only on the very end of the graph ( inner spindle) of the drives where the chipset/drivers/rest of the IO system starts to catch up to the hard drives.  Data access on the inner circles of a hard drive platter of course are slower since the platter always spins at the same speed and of course due to the smaller diameter of the circle, less data is traveling under the head per revolution.  That's why it's only in the very end of the chart where the drives just almost start to become the bottleneck in the I/O system.  In this setup, I could estimate I'm losing at least 80MB/sec due to a lack of controller performance on the motherboard.

Correction: The HD Tune stats of a single 300GB Velociraptor are below...
Transfer Rate Minimum : 72.7 MB/sec
Transfer Rate Maximum : 118.4 MB/sec
Transfer Rate Average : 99.4 MB/sec
Access Time : 7.3 ms
Burst Rate : 158.5 MB/sec
CPU Usage : 2.3% 

More transfer stats on these raptors are reviewed here:

http://www.uberclocker.com/?page_id=95

 So I'm getting only about 60% speed-up when putting two of these drives in RAID-0, mostly due to chipset / driver limitations on the SB700.

 

Now let's take a look at a RAID-5 setup with four hard drives.  3 of the drives are Seagate and the fourth drive is a WD..  Yes, I know it's VERY bad to be mixing drives in a RAID array, but in this situation, I just couldn't get my hands on another old Seagate drive and even if I could, I couldn't stomach paying a premium for it when newer drives give you so much more capacity for your money.   Below is from the RAID Controller's configuration page which is an HTML document that you interact with in a web browser, just like setting up a router..

Again, yes I'm well aware that WDC drive in the RAID array is an ugly duckling and most of the capacity of the WD drive is going to be lost while it's in the RAID-5 array.  The WD drive of course, will shrink down to 149.05GB size of the other drives..  However, as it turns out, this is much less of a problem than you would expect.. In fact, I didn't have any issues at all with the WD drive in the array.  No issues deploying the RAID and certainly, as you will see below, no issues with RAID-5 performance.

 Four Desktop grade 7200 RPM drives running in RAID-5 on a 3Ware card with 256KB stripes.

 

 And there we have the beautiful curve that we would expect to see when the hard drives themselves are the bottleneck in the I/O system.   Also a thing to pay attention to is the CPU usage.  It's only at 3.4%, this is an old AMD Athlon 64 (single core) 2.1 Ghz CPU and it's only using 3.4% of a single CPU core to hit much higher transfer speeds than the setup above.

 

Thanks for reading! Remember people, Raid-5 is not for onboard chipsets... I saw massive performance drops switching from a 3 drive RAID-0 array to a 4 drive RAID-5 array back when I was using the onboard RAID controller. I threw in this 3ware controller and I'm now seeing performance exactly where it should be when the performance of the drives themselves is the bottleneck.  These are not low performance drives either, they had the biggest cache, fastest RPMs, best warranty, and best access times of just about any Desktop level (Raptors I consider to be Workstation level) hard drives out there.


Conclusions that I've made thus far:

  • The SB700 chipset, and most likely other onboard RAID chipsets are insufficient for a RAID-0 setup using Velocirapters if you wish to get all the performance out of your raptors.
  • RAID-5 should not be used with any onboard chipsets if it's at all possible to avoid.  I saw a massive write performance drop moving from three drive RAID-0 to a four drive RAID-5 on an nvidia 570 Ultra chipset.
  • RAID-5 should have a write-back cache to significantly boost performance. I'm seeing over 500% performance boost on random writes with write-back cache enabled on a RAID-5 array. 
  • During a sudden power loss, having a large write back cache will likely mean more lost data and more corruption.  To counter this, make sure you either have an UPS unit or a battery backup module for your RAID Card.  Personally, I recommend an UPS unit since the battery backup modules for RAID cards are quite costly and are often make/model specific to the RAID card.


Coming Soon:

Raptors with a Raid Card! I just need to decide which RAID card to go with as my other RAID card doesn't have any more spare ports.

Published Wednesday, October 07, 2009 11:05 PM by Sidicas

Comments

No Comments
Anonymous comments are disabled

About Sidicas

Boy/Man

 Home   Forums   Chat   Blogs   Deals   Newsletter   About 

 FAQ   Terms of Use   Privacy Policy   Contact Us 

©2009 Newegg, Inc. All rights reserved.