hard drives & raid - benchmark and compare!

[nickhere]

first check all stripe using hdtach use the large zone and note cpu usage , burst and avg read.

 

i belive it should be as close as posable.

 

next you play with cluster size using windows format.

also i notice the formating of the raid must be done full or your read write may suffer

and use ato to detemie read and write speed on 32 mb zone.

 

also hdscan at hddguru.com

will tell you how fast the raid will read a sector.

you want the sector as low as posable.

i had a few bad spots but refreshing it with spinright clear up that problem.

 

on a 2x wd2500ys drive i find i find 32k gave me 4% cpu 114.5 burst and 101.4 avg read.

across the whole drive.

 

any other suggestion.

nick

[Angry_Games]

Core2Duo E6600

975X/G Infinity

2x1GB OCZ Gold GX PC2-8000

X1900XT ATI

2x 250GB Seagate 7200.10 16MB cache SATA II drives:

 

 

 

 

 

[cler1c]

AMD Athlon 64 3200+ Venice

DFI Lanparty nF4 Expert

2x1GB Mushkin XP4000 991483

2x Nvidia 7900GS

2x 160GB Western Digital Raptor 10,000 16MB cache RAID 0 16k stripe 64k cluster:

 

 

[soundx98]

^^^ Do you mean 64 stripe, 16 cluster?

[cler1c]

^^^ Do you mean 64 stripe, 16 cluster?

 

Nope.

[politbureau]

cler1c - that's some high CPU utilization... Have you tried turning NCQ on and increasing the stripe size? Here's my experience with the same drives.

 

My results have been that with NCQ on, my CPU util is only 2-3% on my eXpert. Average read speeds drop ~5-10MB/s, but this is negligible and unnoticeable compared to the parisitic effect of high CPU utilization - think about playing an MP3 and it eating 11-12% of your CPU time while you try to do other things...

 

I've also played with numerous different stripe sizes (16kb up to 512kb) and found that for gaming and media, 32kb or 64kb stripes offer the best overall performance. With the smaller stripe sizes it takes a lot more I/Os to read large files, and your synchronous performance goes out the window.

 

Also, couple things you may or may not have noticed - make sure you enable write back caching through the driver control panel under Device Manager - mine was disabled by default. I should mention that this increases data loss potential in the case of a power failure... And check to ensure the "spread spectrum clocking" jumper is in place on the drive. Mine came without the jumper in there (it was the ROHS version?), and disabling this gave me about 10% better performance.

 

Great drives tho!

 

Peace.

[cler1c]

Thanks a lot, politbureau. Just the sort of help I hoped I would get from this forum.

[politbureau]

Thanks a lot, politbureau. Just the sort of help I hoped I would get from this forum.

No worries So your sig lists your raptors as the 160GB (Dell) version, but you say 150GB in your post - which version are you running? I assume the 160GB since that HD Tach report shows data all the way to 320GB.

 

On a side note, I find it interesting that WD provides this unit exclusively to Dell, considering it is identical to the 150GB, with the last 10GB 'unlocked.' This last 10GB is short stroked on the 150GB version to increase overall performance specifications for the retail market. This is actually pretty clear just from looking at cler1c's HD Tach readout - the average read score would be higher if those last 20GB where it falls below the 120MB/s threshold were not read.

 

Anyone else read anything more on this?

[cler1c]

Thanks again Politbureau.

 

Based on your recs I turned on NCQ and edited my post to show the new results. Not much improvement in CPU utilization. I have my OS and a bunch of other data on these drives so I'm not about to change the stripe size right now but thanks for the info. You're sure stripe size will make that much difference in CPU utilization?

 

Yeah, I have the Dell Raptors @ 160GB, nice drives for a nice price.

[Booster]

Here is mine, maybe I can improve from this but I need to go and double check what my RAID 0 settings were. All I know that I left it default while creating this stripe.

 

This shows both benchmarks with NCQ and Read caching on and off.

 

 

 

As for this in red is with NCQ and Read caching on while in blue shows both off for both 8mb and 32mb.

 

 

 

Just looking at it shows with it off gives me a big boost in burst speed and lower CPU utilization with both options off but the orginal benchmarks I ran I didn't restart my system if that would make a difference.

[soundx98]

Hard to compare to others Booster

You need to have the "total length" set to 32MB and have "neither" selected under Direct I/O

[Booster]

Oops I didn't see that I had to change the settings, updated my previous post with the correct settings in ATTO.

 

Thanks soundx98

 

Edit*

Just wondering does it matter what partition and size should be benchmarked? In my previous post I have it set to a empty 73gb partition and I just ran it again for my windows C partition.