80 MB/sec not enough on today's SCSI controllers?

[Adcadet]

hey crew,
I'm thinking about getting a Atlas 15K drive in the near future. Is there any reason I should consider replacing my SCSI card - it's an Adaptec 3950 U2B (U2W-LVD, 80 MB/sec, dual channel card). I don't do much (anything?) that needs high STR as far as I know, and I don't know of any SCSI command enhancements that I might benefit from for desktop use (have there been any worth mentioning introduced since the U2W-LVD days?). So is there any point in upgrading my card? And if there were a reason, what should I go with these days for reliable yet cheap use?

Thanks!
Andy

 

[The JoJo]

I'm hesitant to recommend an upgrade, I don't think it's worth the dollars due to the small amount of performance you might gain.

...Unless of course you find a very good deal on a 320 controller. I dare say LSI would probably be recommended by many as the manufacturer to look for, due to it's good price/performance.

Ok, now it's time for you people in the US that know the price situation there to speak up!

 

[ddrueding]

Even if there were a need for >80MB/sec transfer rates; I assume you're using a 32/33 PCI bus? You wouldn't gain much regardless. There must be a better place to stick that money...

 

[Fushigi]

LSI chipsets are on at least some of the SCSI cards in my AS/400s. I don't think I could make a better recommendation.

SR's current 15K leaders haven't hit 80MB/s (although they're close). I doubt the new generation will be so much faster that it'll exceed 80MB/s by a great deal. Probably not worth the upgrade unless you'll be running multiple drives off the controller.

 

[CityK]

SR's current 15K leaders haven't hit 80MB/s (although they're close). I doubt the new generation will be so much faster that it'll exceed 80MB/s by a great deal.

For illustrative purposes, the Maxtor Atlas 10K V apparently provides 89-52MBps

 

[Fushigi]

For me, that's not a large enough improvement to justify spending the extra $ of a new controller & cable. Especially since Adcadet said STR isn't as important for what he does. Unless the controller was sufficiently cheap, of course. Depends on Adcadet's budget.

 

[CityK]

I'm in complete agreement. My post was to further highlight the fact that the next generation disks are not, as you said, a great deal faster then 80MBps.

 

[Adcadet]

For me, that's not a large enough improvement to justify spending the extra $ of a new controller & cable. Especially since Adcadet said STR isn't as important for what he does. Unless the controller was sufficiently cheap, of course. Depends on Adcadet's budget.

Can anybody comment with authority (or at least informed opinion) on my lack of need for higher STR (which I pretty much believe already), or any of the newer SCSI features since the U2W-LVD standard?

 

[Mercutio]

Andy: You don't need more STR. Almost no one needs more STR. Unless you're multiplexing realtime AV streams or spend most of your day moving CD Images between drives, you probably don't, either.

High STR helps your OS start a touch faster. Sometimes games load more quickly. If you spend hours opening and closing 15MB photoshop files for fun, maybe. Chances are you'll never notice.

95% of what every normal desktop user does is seek-intensive if it's anything intensive.

 

[Tea]

Well, I can comment with an uninformed opinion, is that near enough?

I run three X15s, two GEN 1 and a 36LP. (All 18GB.) Not the very latest, but not all that old either. The 36LP is on a 160MB/sec controller (Tekram), one of the GEN 1s is on a Tekram 320, and the other one on an Adaptec 40MB/sec controller.

There is a difference between them best described with one of Tannin's favourite phrases, "bugger all".

The 36LP is marginally faster. There was zero difference between the 36LP on a 320 and and on the 160. The 40 holds it back a tad, but not all that much.

STR really is a very minor factor in general purpose use. Access time is what it's all about. But we already knew that.

Now are there any other questions you'd like answered with a complete lack of knowledge but a dash of charm and loads of confidence? Feel free to ask.

 

[Adcadet]

For me, that's not a large enough improvement to justify spending the extra $ of a new controller & cable. Especially since Adcadet said STR isn't as important for what he does. Unless the controller was sufficiently cheap, of course. Depends on Adcadet's budget.

Can anybody comment with authority (or at least informed opinion) on my lack of need for higher STR (which I pretty much believe already), or any of the newer SCSI features since the U2W-LVD standard?

Alright already! Forget STR. Check. Got it. Roger that. But is there any meaninful difference between the older U2W-LVD standard and the newer U320 for desktop use? After looking at SR's TCQ data I'm fairly convinced that TCQ may be nearly worthless for me, but are there any other features added to U320 since U2W-LVD that would help? Or has the U2W-LVD->U160->U320 only involved an increase in bandwidth?

 

[Mercutio]

Didja notice a big difference with your Ultra133 controller on that Athlon64 from the Ultra100 on your old Tyan?

Really, all that U320 added were pretty much limited to datacenter level technologies. The workstation you're using at home does NOT need the refinements made to domain validation or bus arbitration that characterize the U320 spec. Maybe packetized SCSI commands are useful for the day you decide to plug in four X15s but until then, I don't know what you're getting out of that. Essentially everything that was done to U320 in some way improves data transfer rates. I just read a white paper that suggests that U320 SCSI improves recoverability in the event of a system crash, but the paper does not mention any particular technology as being responsible for that, and I can't find anything else about it.

 

[Bozo]

A couple of years ago I built 13 file servers. The files were small, but there was a lot of I/O. Most files needed to be sent across a 100Mb network that could easily be saturated with the latest IDE drive. We wanted redundancy, but didin't need tons of hard drive space.
I used 18GB Cheetah X15s in RAID 1 on an Adaptec Ultra 160 controller.
I have tested two X15s on an Ultra 320 controller and could find no difference.
Access time is the key. That's the only reason I chose the X15s. I could have saved a bunch by buying 10K RPM drives with much slower access times.

Bozo :mrgrn:

 

[Bozo]

One other note: The SCSI controllers were plugged into standard 32 bit PCI slots.


Bozo :mrgrn:

 

[Santilli]

95% of what every normal desktop user does is seek-intensive if it's anything intensive.

Not according to storagereview.com...
:mrgrn:

 

[Adcadet]

95% of what every normal desktop user does is seek-intensive if it's anything intensive.

Not according to storagereview.com...
:mrgrn:

Is it me, or has SR flip-flopped (hate to use a term used recently by politicians) in recent years over this. IIRC, their latest thinking is that larger caches can make up for less than stellar seeks (why the 8 MB drives do so well). But then they claim the Raptors are so great, even compared to other 8 MB drives, so maybe seek is important to SR. Have I read this right, or have I been too involved with other things (hmmm...getting married, school)?

 

[Bozo]

One other thing to consider: 95% of all desktop users never defrag their hard drives. The better the seek time, the less impact this has on performance loss.


Bozo :mrgrn:

 

[Santilli]

WOW! I DEFRAG MY DISKS EVERY SUNDAY... :eekers:

Funny how raptors became the bomb, but I thought it wasn't seek times, but it's cache an algorithim design for using the cache, that allowed it to score wo well in iPeak?

Wonder if you could design a test suite for IPeak that would favor ATA characteristics over SCSI, or one that favors certain drive characteristics?

Boy, I am such a cynic...

:wink:

s

 

[Handruin]

I defrag once a month. More than that causes too much loss of use for my machine (I don't automate this action). I never notice a performance difference anyways, unless if I used a benchmark. At the rate I copy & delete files, I should defrag every few days, but this isn’t realistic. Most of my file copy/delete occurs on a non-boot drive, so system performance doesn't really suffer. When I do defrag, my boot drive is usually the least affected. My media drive is usually 95% red.

Also, that's not really being a cynic; why not cater to your audience (i.e. hard drive)? We do it for people, why can't we do it for a more systematical device such as a hard drive? If you know how to exploit the good in a device, then why not try?

 

[LunarMist]

I used a 15k.3 for nearly a year on an U2W controller then switched to U160 due to controller problems. In every test and active benchmark on a 32-bit PCI bus there was no noticeable difference other than burst rate. I think it will take the next generation (H2 2004) drives to make a difference, amd even that will be insignificant for most uses.

 

[Bozo]

Where I work, we have more than 300 computers. Because it's not automated, I'd bet that 99% of them have not been defragged in 2 years.
On the occasion that I'm ask to look at a computer that is performing poorly (I'm not in IT) I find letting defrag run over night (it usually takes that long) does wonders.

Most home users and office users never defrag their hard drive. Most don't have a clue ( about defragging)

Bozo :mrgrn:

 

[mubs]

A new XP system is terribly fragged to begin with - I don't understand why. The very first defrag of a new system takes ages.

 

[Handruin]

It's not just XP, windows 2000 server is the same way. I can't say if it's worse than XP, because I've never actually compared numbers, but the file system is fragmented enough to warrant a defrag. I usually do this because we make ghost images at work. When I create a new install, I usually defrag before ghosting figuring ghost will capture the organized file system.

 

[Tannin]

I practically never defrag. Defragging makes bugger-all difference to my machines. I suspect that this is because of my normal usage pattern: Install OS; install the apps that I'm likely to use (these rarely change: 95% of what I'll do on a system during the 2 years or so that it's likely to maintain a particlar OS install is run the apps I whacked on the same day I formatted the partition); copy the data files over. Then use it. My data files are almost entirely restricted to seperate physical partitions from my OS install. Most of them are JPGs of medium size, and the great bulk of them are added but not edited.

Anyway, for whatever reason, defragging doesn't seem to make much difference. Though it's six months or so since I installed last, so I must be nearly due. Any day now. Certainly before Christmas.

 

[Handruin]

Are those who constantly defrag, using a file system other than NTFS? Maybe this is part of the reason why Tannin and myself do not notice a difference after defraging? I know I said I defrag once a month, but I have no actual schedule. If I remember, I'll defrag, and it's usually every month or so.

 

[mubs]

My defrag schedule is also on a manually variable basis, usually once every 3 or 4 weeks. Having an older system, there is a noticeable improvement for a while, mainly in booting (every day). I'm using W2k and NTFS. I shutdown every night.

Over the last 6 months or so, booting has become painfully slow in the last step. I'll get the login window in fairly ok time, will put in my password, and the system will sit at a completely blank desktop (with mouse cursor in the exact center) for about 25-30 seconds before all the icons show up. I suspect it was an update to that pox, NAV, that is doing something, that is causing the slowdown.

 

[CityK]

MVB (manual variable basis) as well. I notice a distinct, but short lived, improvement for opening up Explorer. Booting doesn't seem to make a difference...but then again, when I boot up (W2K), I usually go get a fresh pot of coffee going or empty the dishwasher or do something else of remote necessity. In regards to booting, it is certainly one area for which I much prefer XP

mubs, I actually had the opposite experience - something had caused W2K SP3 to take forever to shutdown. When SP4 came out and I upgraded to it, the problem was resolved.

 

[sechs]

I have PerfectDisk set to run every night. Overkill, yes -- but I never get a serious hit from fragmentation, and the disks aren't really doing anything else at two in the morning.

 

[Buck]

I have PerfectDisk set to run every night. Overkill, yes -- but I never get a serious hit from fragmentation, and the disks aren't really doing anything else at two in the morning.

Ditto . . . and this is scheduled shortly after an automated session of Disk Cleanup. No need to defrag files I don't want.

 

[sechs]

I feel that leaving the disk messy gives the program something to do 8)