Sign in to follow this  
honold

3ware 7000-2 + 2xwd2000jb mirror, wb99 still sucks bad

Recommended Posts

business disk winmark = 7370

high-end disk winmark = 24300

this is a clean xp sp1 install as of last night with all windows updates, no tsrs

abit th7-ii motherboard

antec truepower 550w psu

p4 1.8a w/alpha 8942/panaflo high hsf (oc to 2.4, tests same w/o oc)

1gb pc800 rdram - no page file

lite-on ltr-163 dvd-rom and lite-on 32x cdrw on mb ide channels

2xwd2000jb oem with antec cobra round cables (raid 1)

abit siluro geforce4 ti4200 (ti4600 oc)

turtle beach santa cruz audio

intel pro/100s ethernet

3ware 7000-2 pci raid (latest bios/firmware/driver)

3xenermax adjustable 80mm fans (adequate case cooling)

those should be the only remotely relevant details. what could possibly be the deal here?

Share this post


Link to post
Share on other sites
Guest russofris

Honold,

Can you post your Disk Inspection Test graph, I wanna see how messy it is, and whether it's a sawtooth or bandwidth related issue.

I assume that you've tried all the basics (Different PCI slot and such).

Frank

Share this post


Link to post
Share on other sites

just tried another pci slot with strange results - degraded array. it wouldn't see the disk on port1. i popped it back in the other slot, and it detected it. i'm rebuilding the mirror now.

my str graphs are fine, which is one of the many reasons this is so confusing. once the mirror rebuilds i'm going to try the slot again incase i bumped a cable or something.

Share this post


Link to post
Share on other sites

For 3ware 7000-2, i've tried cuda-iv, maxtor d740x & wd800jb, by benching atto @raid-0 all will cap @13+mb/sec write, read is ok. Initially i thought only wd800jb, so i try the two mentioned hdd, d740x & cuda-iv, still the same. As for wb99, the business disk winmark & high-end disk for raid-0 for the mentioned 3 brand of hdd never be good even worst than a single cuda-iv running on a on-board hpt controller & only on par w/ on board via ide controller, ( for benching of wb99 only ) .

Actually what i don't like is under pure dos for ghosting, the transfer rate is simply too slow for the 3 type of hdd, approximately 80mb/min is really too unacceptable for via ide controller to 3ware 7000-2 ide controller, for via ide ~ hpt ide it take approximately 400mb/min for wd800jb, maxtor d740x & cuda-iv, approximately 5x slower, so i think not really a winxp bug though it's slower for basic disk.:lol:

Anyway, all you need is use dynamic disk for winxp, but for me i need pure dos, fast partition dump & load so 3ware 7000-2 is simply not for me, unless new firmware is out that can take advantage of pure dos, i mean fast tranfer rate.:lol:

3k3.

Share this post


Link to post
Share on other sites

it should be observed i had identical performance on a single wd2000jb with the onboard (intel i850) ide and a promise ata133 controller. i've added a disk and began using the 3ware 7000-2, but the performance remains the same :|

Share this post


Link to post
Share on other sites

honold,

How many runs are you doing? Are you rebooting between runs? (see the setting options). SR does nine runs and discards the first two.

Try running without the Lite-Ons attached - I recall something in the past about some of the drives having DMA issues which negatively impacted HDD access times, which in turn would affect your Winmarks. It would make sense cause your office scores (heavily access bound) are 45% of the SR results, and your H/E - although still a dismal 66% of SR's - aren't impacted as heavily, and wouldn't expected to be so because of the more sequentially favoured nature of the test.

Last resort - try installing just XP - no service pack. Although highly unlikely such a major contributer, you still may very well be experiencing scaling differences from the service pack - this can have a significant affect with winbench scores.

CK

CK

Share this post


Link to post
Share on other sites

i'm doing multiple runs with reboots in-between

also, i was utterly wrong about the sawtoothing :( on previous controllers str was fine, but it's looking bad here. i'm posting 'select the drive and click start' atto scores here because wb99 takes forever to run, and these (c = 40gb, f = 140gb, both ntfs) seem to illustrate the point:

atto_c.JPG

atto_f.JPG

Share this post


Link to post
Share on other sites

well here's a dramatic one...

i installed the intel app accellerator and the intel i850 chipset drivers, and here are the new (apps running, unscientific) attos:

atto_c2.JPG

atto_f2.JPG

Share this post


Link to post
Share on other sites

i just unplugged drive 1 (my original drive) to see if the other one has the problem, and it has the exact same results.

while my drives do have close serials and the same firmware, they were bought a month apart from different vendors. i seriously doubt this is a hardware issue.

Share this post


Link to post
Share on other sites
it takes approximately 25 seconds to copy a 440mb file on c: to a temp file on c:
That's normal. But, if you copy from other physical drive to C, then you will get half the time it takes to copy from C to C.

Share this post


Link to post
Share on other sites
Actually what i don't like is under pure dos for ghosting, the transfer rate is simply too slow for the 3 type of hdd, approximately 80mb/min is really too unacceptable for via ide controller to 3ware 7000-2 ide controller, for via ide ~ hpt ide it take approximately 400mb/min for wd800jb, maxtor d740x & cuda-iv, approximately 5x slower, so i think not really a winxp bug though it's slower for basic disk.:lol:

3k3.

Your system definately needs some fixing. It's not either controller or hdd.

Share this post


Link to post
Share on other sites
atto, winbench 99, and sandra scores are off.  i somehow doubt i have a 'benchmark bug' :(

those are all low level benchmarks.

wd pata drives are not very good raid drives to begin with; however in order to bench your setup properly you need something that works through the miniport driver rather than direct io.

when you have any redundancy on a high level io switching controller like 3ware etc., the priority of the asic as well as the miniport is a safe write.

quite a few run of the mill controllers do not pay attention to little things like that and just do WB type of operation for any raid level.

3ware's 3dm/miniport is tuned so that there's a secured write after each io. you can play with the miniport as well as 3dm to tip the scales toward faster vs reliable, but would that be the purpose for setting up raid1?

Share this post


Link to post
Share on other sites
it should be observed i had identical performance on a single wd2000jb with the onboard (intel i850) ide and a promise ata133 controller.  i've added a disk and began using the 3ware 7000-2, but the performance remains the same :|

Share this post


Link to post
Share on other sites
it should be observed i had identical performance on a single wd2000jb with the onboard (intel i850) ide and a promise ata133 controller.  i've added a disk and began using the 3ware 7000-2, but the performance remains the same :|

RAID1 doesn't give any performance increase. Basically it just reads from one disk and writes to two disks. 3ware claims to have technology called "TwinStor", which should allow simutaneous reads. I haven't seen any benchmarks that would proof this though.

Jan

Share this post


Link to post
Share on other sites
RAID1 doesn't give any performance increase. Basically it just reads from one disk and writes to two disks. 3ware claims to have technology called "TwinStor", which should allow simutaneous reads. I haven't seen any benchmarks that would proof this though.

One need look no further the StorageReview STR results for the 3ware cards. For two disks the RAID1 read STR is about 2x the read STR for one disk. The write STR is a little below the write STR for one disk.

My HighPoint 404 controller does the same thing. Read STR for a RAID1 is double the read STR for one disk. So the RAID1 STR is roughly equivalent to the RAID0 STR when reading, and to the single-disk STR when writing.

Share this post


Link to post
Share on other sites
RAID1 doesn't give any performance increase. Basically it just reads from one disk and writes to two disks. 3ware claims to have technology called "TwinStor", which should allow simutaneous reads. I haven't seen any benchmarks that would proof this though.

One need look no further the StorageReview STR results for the 3ware cards. For two disks the RAID1 read STR is about 2x the read STR for one disk. The write STR is a little below the write STR for one disk.

My HighPoint 404 controller does the same thing. Read STR for a RAID1 is double the read STR for one disk. So the RAID1 STR is roughly equivalent to the RAID0 STR when reading, and to the single-disk STR when writing.

Damn... I think my crappy integrated Promise FT133 "Lite" doesn't do that... Well, as Eugene has said it, STR has little effect in disk performance. I'd still like it to fly. Do You know whether HPT370 could do "load balancing"? I have one lying useless.

Cheers,

Jan

Share this post


Link to post
Share on other sites
RAID1 doesn't give any performance increase. Basically it just reads from one disk and writes to two disks. 3ware claims to have technology called "TwinStor", which should allow simutaneous reads. I haven't seen any benchmarks that would proof this though

i am getting terrible metrics in both single drive (3 different controllers) and raid 1. i am not whining about raid 1 not giving me a performance boost.

Share this post


Link to post
Share on other sites

honold,

if you have any other drives to play with, check them out. as i have mentioned above wd drives do not play well in raid. they are good desktop drives when you do just one drive per channel and don't try anything funny. you can also confirm this by doing a search on this board for something like "3ware poor performance"

i've always had best luck with ibm drives as they are the most advanced circuitry/command set wise. if you call 3ware, they will recommend seagate, maxtor or ibm as well. most ata raid quals are done with ibm drives as well as they really know how to use tag and command queing and have a good seek time compared to seagate.

Share this post


Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
Sign in to follow this