Why not SCSI?

[Caecus Veritas]

hm... i thought one of the main advantages of SCSI drives, whether raided or single, was their capability for sustained ... err what was the term... data transfer rate (i don't think this is the correct terminology, but anyhow), whereas IDE drives used bursts of data packets. this still holds true, no?

 

[toyotabedzrock]

SCSI in general can handle more concurrent IO operations, and at the same time in a non raid setup will use almost no additional CPU cycles for those extra IO ops. Also a 15k drive provides a faster random access time. So if you run allot of things at the same time you will feel the added performance.

Also the reason SCSI drives provide less storage is because they are designed for business usage, in other words there data is worth allot of money. You can also put 15 of them on a single SCSI channel, again another buisness use.

::side note::
I own a second generation still, its far to slow now days for most things but i left it in my parents computer to hold apache and mysql for testing web apps at home. Also it has run basically 24/7 for about 10years now. Admittedly the last last 5 years is has done nothing but light file/web/sql serving, and holds a swap file. But its still a nice long life. They do run hot and are loud. Original price was about $1200 for a single drive ~4GB.

 

[bfonnes]

Originally posted by: Fullmetal Chocobo
At this point, you'd be better off at looking for SAS hardware.

What does SAS stand for? Stinks As SCSI?

BFonnes

 

[GarfieldtheCat]

SAS = Serial Attached SCSI

Link to Single-User Benchmarks

Link to Gaming Benchmarks

150Gig Raptor wins against current model 15K SCSI drives. No reaason now to get SCSI for single user use nowadays.

 

[bob4432]

Originally posted by: GarfieldtheCat
SAS = Serial Attached SCSI

Link to Single-User Benchmarks

Link to Gaming Benchmarks

150Gig Raptor wins against current model 15K SCSI drives. No reaason now to get SCSI for single user use nowadays.

current gen fujitsu are max/may(sas) series, those are mau series...but i haven't seen benches on the max's/may's so i can't comment

 

[SunSamurai]

Originally posted by: Fullmetal Chocobo
At this point, you'd be better off at looking for SAS hardware.

Dont you SAS me!


Sorry had to.

 

[OSX]

Right now, SATA takes the easy market, which is the single user stuff. SCSI, and now SAS have always outperformed everything else in heavy use multiuser environments. If you're building a gaming rig, get a Raptor, if you're building a database server, you're an idiot if you don't have 15K SCSI drives in there. Each interface has it's own market.

 

[OvErHeAtInG]

Originally posted by: OSX
Right now, SATA takes the easy market, which is the single user stuff. SCSI, and now SAS have always outperformed everything else in heavy use multiuser environments. If you're building a gaming rig, get a Raptor, if you're building a database server, you're an idiot if you don't have 15K SCSI drives in there. Each interface has it's own market.

QFT.

Also, the SCSI drives I've used have always sounded roughly like a bandsaw. I know that doesn't apply to all of them, but most servers I hear -- I just wouldn't want that whine on my desk.

 

[toyotabedzrock]

I think you guys should look at the other benchmark results.

Random Access Times

Also i see something very important in those single user benchmarks, the NCQ is turned off for the raptor.

In the gaming benchmark the Cheetah outperforms the raptor in Far Cry and World of Warcraft. But falls behind in the Sims, not the most demanding game.

Multiuser Performance

Here the raptor is falls way short of even the 10K scsi drives.

 

[bob4432]

Originally posted by: toyotabedzrock
I think you guys should look at the other benchmark results.

Random Access Times

Also i see something very important in those single user benchmarks, the NCQ is turned off for the raptor.

In the gaming benchmark the Cheetah outperforms the raptor in Far Cry and World of Warcraft. But falls behind in the Sims, not the most demanding game.

Multiuser Performance

Here the raptor is falls way short of even the 10K scsi drives.

NCQ on the sata drives has shown to actually hinder performance on some drives and definately is not as matures the scsi setup.

when you compare a 10 & 15K of the same size format (3.5") drive for random access, it is just pure rotational speed that always wins.

also, when you guys speak of the whine on the 15k drives, what drives do you speak of? maybe i am lucky and can't hear that particular frequency very well because the ones i have had don't bother me in the least, at most sound like a noisy 7.2K hdd, nothing more.

 

[GarfieldtheCat]

Originally posted by: bob4432

Originally posted by: GarfieldtheCat
SAS = Serial Attached SCSI

Link to Single-User Benchmarks

Link to Gaming Benchmarks

150Gig Raptor wins against current model 15K SCSI drives. No reaason now to get SCSI for single user use nowadays.

current gen fujitsu are max/may(sas) series, those are mau series...but i haven't seen benches on the max's/may's so i can't comment

You are right, and I haven't seen any benchmarks either.

But I thought that the physical specs of the MAX series were about the same to the MAU (again, I think, I'n not 100% sure).

 

[GarfieldtheCat]

Originally posted by: toyotabedzrock
I think you guys should look at the other benchmark results.

Random Access Times

Also i see something very important in those single user benchmarks, the NCQ is turned off for the raptor.

In the gaming benchmark the Cheetah outperforms the raptor in Far Cry and World of Warcraft. But falls behind in the Sims, not the most demanding game.

Multiuser Performance

Here the raptor is falls way short of even the 10K scsi drives.

access time doesn't always mean that much. How the disk does it's seeks (which is based on the HD firmware) and cache size both help. That is why the Raptor wins the single user tests in that review. It's similar to saying the Pentium 4 runs faster (more GHz) then the Athlon64, but in gaming benchmarks, the Athlon64 is faster (does more work).

And yes, in ,multiuser use (I.E. servers), SCSI wins easily. But for a person using a PC as his/her gaming/workstation/general PC, that is not multi-user use, and thus the raptor is faster for them.

 

[Rubycon]

A Word of Caution to Power Users

It is all too common for an enthusiast to believe that his or her usage pattern is closer to that of a server's rather than a desktop's. This idea arises from a variety of sources- "I multitask a lot," "I hear the hard drive grinding away," "I deal with lots of huge files," etc. The truth is, however, that even the heaviest, grinding multitasker experiences disk access patterns that are far more localized in nature than the truly random access that servers undergo. Individuals who choose a hard drive based on its prowess in IOMeter with the belief that their usage habits mimic a server simply do themselves a disservice. It is measures such as the SR Office and High-End DriveMarks that most accurately depict a non-server's response, whether it be the sheer speed experienced under intense disk access or the "snap and feel" associated with intermittent but bursty operations.

That pretty much sums it up.

 

[BoberFett]

Originally posted by: MS Dawn

A Word of Caution to Power Users

It is all too common for an enthusiast to believe that his or her usage pattern is closer to that of a server's rather than a desktop's. This idea arises from a variety of sources- "I multitask a lot," "I hear the hard drive grinding away," "I deal with lots of huge files," etc. The truth is, however, that even the heaviest, grinding multitasker experiences disk access patterns that are far more localized in nature than the truly random access that servers undergo. Individuals who choose a hard drive based on its prowess in IOMeter with the belief that their usage habits mimic a server simply do themselves a disservice. It is measures such as the SR Office and High-End DriveMarks that most accurately depict a non-server's response, whether it be the sheer speed experienced under intense disk access or the "snap and feel" associated with intermittent but bursty operations.

That pretty much sums it up.

Depends on the usage. Modern MMORPGs have several GB worth of data (I think EQ2 is like 5GB) and updates are done frequently with small pieces of data. Depending on how often a person who plays such a game defrags, that data can be all over the hard drive, and accessed randomly. Because you never know what model you're going to run into and what textures might be needed, the data could be anywhere. Outside of zone data, it's hard to predict and compile data so that it's arranged logically. In that type of situation SCSI may have a distinct advantage. I've noticed far more stuttering on my wife's computer than mine while running through a town in EQ2, she has a 7200 IDE drive, I have a 15K SCSI.

 

[Chaotic42]

Good SCSI drives are epxensive as others have said. I have an LSI-U160 controller in a regular PCI slot with a Fujitsu MAU 15K 36GB drive. According to HD-Tach it beats my roomate's new 150GB Raptor in sustained read rate. I get about 81MB/s sustained.

I like SCSI, but it really is expensive. That and the fact that there aren't any user-level PCIe SCSI cards out there. I can't find one for less than about $600.

 

[toyotabedzrock]

Originally posted by: MS Dawn

A Word of Caution to Power Users

It is all too common for an enthusiast to believe that his or her usage pattern is closer to that of a server's rather than a desktop's. This idea arises from a variety of sources- "I multitask a lot," "I hear the hard drive grinding away," "I deal with lots of huge files," etc. The truth is, however, that even the heaviest, grinding multitasker experiences disk access patterns that are far more localized in nature than the truly random access that servers undergo. Individuals who choose a hard drive based on its prowess in IOMeter with the belief that their usage habits mimic a server simply do themselves a disservice. It is measures such as the SR Office and High-End DriveMarks that most accurately depict a non-server's response, whether it be the sheer speed experienced under intense disk access or the "snap and feel" associated with intermittent but bursty operations.

That pretty much sums it up.

It still wins in the game benchmarks, and the second test is called multiuser not server again the SCSI drives win there. And there is a reason desktop systems are moving to multiprocessor type platforms.

Also in pure transfer rate and access time the 15k drives are better.

Side not segate has a utility to tune its drives a lil more for a desktop environment. Also will point out that if you buy a workstation type machine that is more of a high end machine they most times include a high speed SCSI drive.

 

[Pariah]

Originally posted by: bob4432

current gen fujitsu are max/may(sas) series, those are mau series...but i haven't seen benches on the max's/may's so i can't comment

The "current" gen Fujitsu are simply RoHS compliant versions of the old drives. Performance is the same. Also, the current gen drives are MAX and MAW. MAY is the 2.5" form factor 10k competing with Seagate's Savvio.


There is no one drive wins everything drive anymore, but for home users, no matter what kind of ridiculous multitasking you think you do, the 150GB Raptor is the overall performance champ. When considering it's also a price/capacity winner vs competing SCSI products, it's really a no brainer. The 15k.5 may change that, but it isn't supposed to make an appearance until October at soonest. The debate used to be, do you want to pay a premium for the premium performance SCSI gives. Now you don't have to, with the current Raptor you can have your cake and eat it too.

Also the reason SCSI drives provide less storage is because they are designed for business usage, in other words there data is worth allot of money.

No, it isn't.

 

[GarfieldtheCat]

It still wins in the game benchmarks, and the second test is called multiuser not server again the SCSI drives win there. And there is a reason desktop systems are moving to multiprocessor type platforms.

Multiuser = more then one person, not more then one processor.

For a single user, the raptor is still just as fast or faster then current 15K SCSI drives. A single user that multitasks, is still a single user. You will not notice the difference in a blind test between a raptor and 15K SCSI.

SCSI historically was used in business because they are faster for server (ie multiuser) use, and are built to a higher reliablility standard. (5 year warranty).

Unitl recently, yes, SCSI was faster on the desktop then IDE/ATA, but know with the raptor, it has caught up and there is no performance difference.

 

[Ayah]

http://www.newegg.com/Product/Product.asp?Item=N82E16816118027

SCSI PCI-E controller by LSI. Only 649.99!

 

[blackangst1]

Originally posted by: GarfieldtheCat

It still wins in the game benchmarks, and the second test is called multiuser not server again the SCSI drives win there. And there is a reason desktop systems are moving to multiprocessor type platforms.

Multiuser = more then one person, not more then one processor.

For a single user, the raptor is still just as fast or faster then current 15K SCSI drives. A single user that multitasks, is still a single user. You will not notice the difference in a blind test between a raptor and 15K SCSI.

SCSI historically was used in business because they are faster for server (ie multiuser) use, and are built to a higher reliablility standard. (5 year warranty).

Unitl recently, yes, SCSI was faster on the desktop then IDE/ATA, but know with the raptor, it has caught up and there is no performance difference.

Sorry to disagree with you sir, but you are wrong. In single applications, Raptor *may* be equal to SCSI, but if you do ANYTHING that involves hard drive usage with more than one app, SCSI beats it far and away. SCSI is still the only HD that handle multiple threads, simultaniously. This is the reason I sold MY raptor and went back to SCSI. Raptor is just too bursty for me.

 

[spike spiegal]

but if you do ANYTHING that involves hard drive usage with more than one app, SCSI beats it far and away

Bull. SCSI controllers are not aware of the applications you are running and are not 'threaded', so please dispense with the billiard ball physics and glorification of SCSI.

SCSI controllers are capable of far superior command queuing over your typical onboard SATA or IDE controller. This means if your I/O consists of a lot of small data chunks then SCSI is going to show an improvement with random seeks over IDE/SATA.

I have servers right now running both 15K SCSI and vanilla 7200rpm SATA drives on both high end controllers like 3Ware and cheap RAID cards. The justification being the price of 400gig of fault tolerant SATA storage -vs- 400gig of fault tolerant SCSI is enormous.

The SCSI drives are quite a bit faster at writing data, and better at highly random seek patterns. The big SATA drives are only a notch slower in terms of data reads, and perform equal in general operations. Otherwise, unless you're running Netware 3.1 with 64meg of RAM I could not honestly justify expensive, power hungry, noisy and often proprietary SCSI storage.

That's really nice you noticed a big difference onyour system, but we have claims on Anandtech that the brand of RAM you use makes a big difference in performance as well.

 

[cleverhandle]

So many people in this thread trying desperately to justify the money they sunk into a 15K SCSI setup...

I used to be like that, then I grew up. Look at the benchmark numbers, understand the principles behind the SR benchmarks, do some blind tests if you're really not convinced. Buying into a 15K setup for a desktop machine is just a stupid waste of money.

 

[Smbu]

Originally posted by: Chaotic42
Good SCSI drives are epxensive as others have said. I have an LSI-U160 controller in a regular PCI slot with a Fujitsu MAU 15K 36GB drive. According to HD-Tach it beats my roomate's new 150GB Raptor in sustained read rate. I get about 81MB/s sustained.

I like SCSI, but it really is expensive. That and the fact that there aren't any user-level PCIe SCSI cards out there. I can't find one for less than about $600.

If you don't mind getting a used card you can pick up some pretty cheap cards off of ebay. I've seen LSI MegaRAID SCSI 320-2E dual channel U320 PCIe Raid cards for between $100-$200(depending on the seller). Some as rebadged Dell Perc 4e/DC cards. I think the retail price for these cards is like +$600. Good for a cheap PCIe U320 Dual Channel RAID card.

I've been thinking of picking up one these cards as all I have now is an old adaptec 29160 that I've had since forever with my Maxtor Atlas 15K II drive and some old 10K drives I've had from awhile back.

 

[d3n]

SCSI 320 is bottlenecked at this point. It has no where to go. The best option right now for end all be all bang for the buck is to go SAS. The SAS protocol can handle regular SATA drives. I wouldn?t go that route over new SAS drives unless you can throw 5 or more SATA drives at it since the combine throughput can make up for the old SATA tech.

I would look at something like this adaptec enclosure with an adaptec controller card that can do RAID 10 or RAID 6. Stuff your internal bays and this thing full of SATA II 500 GB drives and the combinded IO on the SAS backplane will smoke any high-end SCSI setup. A RAID 10 setup would help the random write times allot.

LSI makes some stuff too but it looks like adaptecs is a little more polished right now. I am hoping to get a SAS snapserver 550 for work, SAS on the backend and ISCSI on the front.

 

[L00ker]

The real reason why scsi is preferred in the enterprise market is reliability, to put it simply AFR's (Annual Failure Rate) on a SCSI 15k drive are ~.65% SATA drives are ~.75% so lets extrapolate that a little, so ~.65% of drives fail in a SCSI solution which means for every 100 15k drives 6.5% fail (100 drives in a datacenter is not much vice 7.5% for 100 7.2k RPM SATA drives. I work for a storage company (not EMC) and the only reason we use SATA drives in our arrays is because they are cheaper, for instance a 300gb SATA drive is ~$80-$100 vice the SCSI near equivalent is ~$275-$300 You do the math, for the cost of 1 SCSI spare you can have 3 SATA spares...