SCSI HD Performance Slower Than IDE???

[resinboy]

Hi gang: just finished installing an IBM 9 gig 10K LVD hard drive, on a Tekram DC390 U2W card. Maximum transfer rate for card is 80 mb/s. Drive is rated at 160 mb/s maximum. I have lvd cables. Card is 2 channel: HD is alone on the UW2 channel, I have a burner, and reader on the other channel.
My IDE IBM 75 gxp gave me Sandra scores in the 23,000 range: I just got done benching my new scsi setup, and it's almost 1/2 half of that:12, 250 ( slower than a UDMA66 HD!!!)
DO you think the 80 mb/s of the Tekram is a significant bottleneck??? HD Tach gives me slightly better rates with the scsi!! I just got this card, but if this is slowing me up this much, guess I gotta lay out the clams once again
Please step on up , and give me your opinions!!!

 

[Fandu]

First, I don't think think your benchmarks are near accurate, here is my experience with a similar setup.

What settings are you using in the 390 BIOS? What options do you have enabled, and what transfer modes are you in?

But assuming that everyhting is set correctly, and the benchmarks are accurate, my second guess is that your reading the outer tracks on the IDE and the very inner on the SCSI drive.

 

[resinboy]

OK- had my coffee- here are the actual#'s
IBM 7200 75gxp (15.3 gig) IBM DRVS 9gig scsi 10K

(below #'s are ide/scsi)


HD TACH
Random access 10.9ms / 9.4 ms
Read speed 58.1 / 56.7
cpu utilization 1.4% / 6.5%

Sisoft Sandra

Buffered Read 440 mb/s / 322mb/s
Sequential Write 28 / 17
Random Read 9 / 8
Buffered Write 263 / 222
Sequential Write 13 / 5
Random Write 11 / 6

Tekram Bios Setup

Adapter id 7
boot id 0
Boot device lun 0
Boot from cd disabled
int 13 extension enabled
greater than 1 gig support-enabled
Rem. media as bios device-boot only
Immidiate return on seek command- disabled
scan all lun for devices-enabled
active negation-enabled
delay before bios scan-3sec
max # of tagged commands-32
scam- disabled

Device options

parity check-yes
send start unit command-yes
tagged command queing-yes
wide negotiation-yes
max transfer- 80 mb's

Interestingly, the scsi set-up has a higher cpu utilization- I figures it would be less than the ide unit.

 

[Howard]

Fandu, it's the other way around. Faster seeks in the middle

 

[Soccerman]

first of all, Sisoft Sandra isn't accurate.

second of all, IBM's 75gxp get really good scores because it has a very high transfer rate. I'm guessing Sisoft doesn't take random access into account much..

 

[rawko]

It's a proven fact that with the exact same setup, an adaptec scsi card will get better transfer rates, atleast in hdtach.

 

[Fandu]

I was thinking about the transfer rates when I talked about the tracks there Yucky, but your right, you get faster seeks on the inner tracks.

Rawko, Have you documented such an experience? Because never once have we had a Tekram card perform any worse that an Adaptec, or vise-versa.

Resin, it certianlly looks like you've got everything setup correctly. (You do have the chain terminated right?). That IDE cpu utaliziation score is insanely low, you arn't using a add-in IDE card like a Promise are you?

What do the performance graphs in HD Tach look like?

 

[rawko]

I've tested it out multiple times, with tekram, symbios, and adaptec cards, but the difference between tekram, and adaptec isn't that far. Symbios is the worst IMO.

 

[resinboy]

Fandu: yes, the chain is terminated with a scsi lvd/se terminator.
The performance graphs look like a red squiggeley line that runs from left to right( sorry, had to be a smartass- I don't know how to interpret that particular chunk of data. I assume it's for read/ write at different portions of the disk? And, no I am not using an ad- in card. It's weird, the system seems snappier, boot time is a little quicker, even with the scsi bios screen. I have tried saving a couple of Tiff files to the HD, and it seems mighty fast.
At this point, would going to the 160 mps card make a difference, or would it strictly help burst speeds, which I probably won't touch?

 

[rawko]

A 160 card will only affect your THEORETICAL burst speeds. But, if you do a get a card, get an adaptec . I would recommend the 2940U2W, if you can find it signicantly cheaper than the 29160.

 

[Radboy]

You should ask at storagereview.com bbs. Those guys know all about HDDs. I read some posts there that said today's HDD benchmarking utilities are *sequential*, which means they don't take into account SCSI's multitasking capabilities (which negates SCSI's primary advantage over the IDE interface). As such, they discriminate against SCSI, in favor of IDE's (non-multitasking) interface (vast majority of users have IDE drives, not SCSI).

What drive do u have? DRVS doesn't tell me anything. 9LZX? 18LZX? or 36LZX? From your scores, sounds like u hava 9LZX. If this is the case, ur comparing a 2-yr old SCSI drive w/ current generation IDE. I have both a 9LZX & 18LZX. I get ~13,500 Sandra on my 9LZX & ~20,000 the 18LZX. Don't have a 36LZX, but I'm sure it would score higher than the 18LZX. If ur getting 12,250 on a 9LZX, that sounds about normal. If u have a 36LZX, something's wrong. How much data is on the drive? fragmentation?

Despite the (theorectical) benchmark scores, the multitasking/multithreaded SCSI interface will still do a better job at running ur OS & apps.

Do u have latest bios (April) & drivers (7-5-00) for the Tekram card?

I have always heard that the Symbios chipset is a hi-perf chipset from LSI Logic, and is better/faster than the comparable Adaptec counterpart (which, in this case, would be the AHA2940-U2W). I've read quite a few posts at the storagereview, and no one ever said the Adaptec card was faster than the Tekram (at the LVD level).

Benchmarking numbers are (merely) theorectical. What about real-world perf? Do u notice any improvements in system responsiveness?

It's easy to confuse the 68-pin UW connector w/ the 68-pin LVD connector, cuz they look identical (I made this mistake once). Does the drive show up at 80MB/s max x-fer in the bios scan? I suggest setting the autostart jumper.

What about write cache? All my IBM drives shipped from manufacturer w/ the write cache *disabled* - which will kill perf. I used (Adaptec's) EZSCSI5 to enable the write cache on all my SCSI drives.

This may help.

 

[rawko]

You know, they probably disabled it for a reason, but i would just leave it the way it is. I messed around with it once with adaptec something scsi 4.0 or something like that, and every time i rebooted it got reset, so i figured it was something with the drive defaulting that way anyway.

 

[Radboy]

I called IBM TS. They said they disable it for reasons of 'data integrity' - even tho the other manufacturers like Seagate & Quantum enable it for better performance. Enabling the write cache will allow ur system to move onto to other things as soon as the data is in the cache (ram, faster). When it's disabled, the system has to wait until the data is actually on the disk itself, which, naturally takes (relatively) much more time.

The prob u run into w/ write cache enabled is, if u hava power failure while critical data is in the cache, but b4 it makes it to the disk, u could be screwed. This is why it's prudent to have a small back-up UPS - just incase. This way, if u lose power, you'll be able to manually shutdown ur PC on battery power.

I have the BackUPS 650 by APC.

 

[resinboy]

Radboy:
yup, she's a 9LZX. Drive has about 2/3 data written on it. I am a psycho about defragging, so that's not an issue. I definitely have 80 mps throughput. ( tried both connectors on card, and observed the bios reeding during boot). System definitely seems snappier. Why would your 18 lzx be faster than the 9lzx? I figured with a smaller drive, access times would be slightly faster( silly me, I guess).
No, I have not updated the card's bios, but I do have the latest drivers.

Yes, I am using a Deltec 700 VA back- up.

 

[Radboy]

The 18LZX is a newer, better, faster drive (newer generation). It has access time spec of 4.9ms, while the 9LZX has 5.3ms.

What about ur write cache?

 

[resinboy]

write cache is however it is set at default. I don't have the printout at hand of the drive exterior- is it a jumper setting?

 

[Radboy]

No jumper. Need a utility to WCE. IBM makes one, but it's pain to use. Adaptec's EZSCSI5 is the easiest. u have?

 

[resinboy]

no gots

 

[resinboy]

I just finished flashing the bios( old one was original, from 3/99)

Re-ran HD Tach

(Old figures/ new figures)

Random access 9.4ms/9.2ms
Read Speed 56.7/58.1
CPU util. 6.5%/5.5%

Not drastic, but an improvement none-the-less. It's nice to see an update work correctly.

Sandra scores: old/new

Buffered read: 322/440
Sequential Read 17mps/18mps
Random Read 8/8
Buffered write 222/253
Sequential Write 5/5
Random Write 6/6

Again, slight improvement. These were an average of 3 tests, with 3 different reboots, so as to eliminate any casheing of data.

 

[Radboy]

If u enable ur write cache, u should see Sandra scores jump ~1500 pts.

I have 9LZX partitioned in 2 halves. The first (faster) half (1/2 full of data) scores ~12,500 w/ WC DISabled, & 14,000 w/ WCE. But real-world perf improvements are more dramatic than benches would indicate - especially since u have big, 4MB cache .. cuz most of the file an OS x-fers are less than 4MB.

I noticed more zing after enabling WC .. and since u already have b/u pwr supply ...

After u get ur 9LZX purring, go here & checkmout this badboy. 15Krpm spindle, 3.9ms access. Guys at storagereview.com are raving about it, saying it runs cooler & quieter than 10K drives, and has longest MTBF specs (reliability).

Hypermicro has it for $475 - that's cheaper than I paid for my 1st 9LZX (when it first came out).

 

[Radboy]

Huh?

Tell me how, & what effect it has.

 

[thermite88]

Radboy worte:

<< All my IBM drives shipped from manufacturer w/ the write cache *disabled* - which will kill perf. >>

I totally agree with Radboy. I have had several IBM SCSI drives which all came with the write cache disabled. It was done to assure data integrity because most of these drives were used in server or workstation environments supporting multiple users. It has a huge penalty in performance. For PC use, the first thing to do is to enable the write cache

It used to be a chord to enable the wirte cache. The Adaptec EZSCSI can do it but will not make it stick. It will be disabled when the system reboot next time. IBM provides a ultility program free for technician and VAR, but they are very reluctant to release it to individuals. (The ultility allows the user to change any mode parameters in the hard drive BIOS. It can be very dangerous in the hand of an inexperienced user and ruins the drive completely. I will not pass it to anyone else due to an agreement with IBM.)

However, the Ultra160 series Adaptec host adapter makes it very easy. The user can enable the write cache from the Adaptec BIOS. You do it once and for all. It changes the drive write cache default to enable or disable. The result is a huge increase in drive speed.

 

[resinboy]

just got done using EZSCSI 5 .
I know Sanda is squirelly at times( maybe most) , but after enabling write casheing, I again have revised figures: I will list the last test with new bios, and then write cache enabled.

Buffered Read 440/402 mb/s
Sequential Read 18/17 mb/s
Random Read 8/8 mb/s
Buffered Write 253/412!!! mb/s
Sequential Write 5/12!! mb/s
Random Write 6/10 mb/s
Also,Thermite88, when I rebooted, write cashe was still enabled

 

[Pariah]

&quot;All my IBM drives shipped from manufacturer w/ the write cache *disabled* - which will kill perf&quot;

This isn't true. The drives are not shipped with the write cache disabled, they are shipped with the ability disabled. There is no such thing as write back cache, it uses whatever the quoted cache amount on the drive is. It won't &quot;kill&quot; performance either no matter what some benchmark tells you. Anyone using Sandra as a HD benchmark, has no clue what they are doing, as it is probably one of, if not the worst benchmarking utilities out there. There will be improvements from enabling write back, but you will be hard pressed to tell the difference.

&quot;15Krpm spindle, 3.9ms access.&quot;

Seek, not access.

 

[rawko]

I totally agree.