How much faster are SCSI drives compared to SATA/PATA drives?

I

imzjustplayin

[H]ard|Gawd
Joined
Jan 24, 2006
Messages
1,171
Do SCSI drives still hold the lead that they once did back 10 years ago over PATA/SATA drives? Because PATA drives are starting to get really fast now but I'm wondering if SCSI drives still have the faster transfer rates of the PATA drives or not. I do know they have faster access times but I'm starting to question the Transfer rates.

There is one thing that I've found to be a disturbing trend as that is the seektimes on PATA/SATA drives... They're getting worse... Anyone notice this? I remember seektimes of 8ms on my older desktop HDDs but now I'm seeing 10, 11 and 12ms which bothers me...
 
When did you see 8ms seeks? On what drive? What testing application?
 
Seacrate 7200.10's are rated for 11ms average; that's sequential track seek, I'll add.
15ms random out of that is nothing short of miraculous, and probably spending a LOT of time in cache.

Now compare to my average <4ms random on old 10000RPM Fuji U320's.

If you want seek speed, stay the hell away from perpendicular. That's all there is to it. Great for capacity, but seeking sucks and then some - don't expect it to improve any time soon either. Seagate's always had crap seek algos.
 
depends on the following:
  • the drives that are compared
  • the application under investigation, i.e. the access pattern
  • the ``drive arrangement'' and controller, if the comparison is not between two single drives
 
The Raptor 150 typically has average seek times of 8ms according to current benchmarks. The just posted Raptor 150 vs the RAID SATA drives over at THG showed 8 to 8.5 ms average seeks for the Raptor compared to 13 to 13.5 ms average seeks for the RAID drives (2x400 WD drives, iirc).

SCSI drives have fallen by the wayside over the years because nowadays with the Raptors you can get nearly the same level of performance at typically a lower cost overall. It's almost impossible to find a mobo nowadays that doesn't have SATA controllers on it; can't say that for SCSI support directly on mobos, and if they do have it, the price is going to reflect it.

So, in the end, the TCO (Total Cost of Ownership) for a Raptor powered machine (even a 2x150 RAID Raptor setup) would still be significantly less than owning a similarly configured machine using a single 15K SCSI setup. And the Raptors will either perform just as well or they might outperform the 15K SCSI setup - and for the lower cost.
 
bbz_Ghost said:
The Raptor 150 typically has average seek times of 8ms according to current benchmarks. The just posted Raptor 150 vs the RAID SATA drives over at THG showed 8 to 8.5 ms average seeks for the Raptor compared to 13 to 13.5 ms average seeks for the RAID drives (2x400 WD drives, iirc).
Click to expand...

Thanks for instantly disqualifying your arguments!

SCSI drives have fallen by the wayside over the years because nowadays with the Raptors you can get nearly the same level of performance at typically a lower cost overall. It's almost impossible to find a mobo nowadays that doesn't have SATA controllers on it; can't say that for SCSI support directly on mobos, and if they do have it, the price is going to reflect it.
Click to expand...

Bzzt, and thanks for playing. You already disqualified yourself, so you don't get to spout more shill from Tom's!
FACT: SCSI has and always will seek faster. Period. Faster RPMs, better algorithms, and superior command queuing.
FACT: SATA's only savior is it's huge buffers and the bell curve for such being around 20MB. A 10000RPM 16MB Hitachi FC-AL will happily rape a Raptor, then go for the money shot while the Raptor's still seeking.
FACT: SCSI doesn't ship on motherboards because your choices for integration are LSI (expensive) and Adaptec (garbage.) But you've NEVER had SCSI shipping on anything except high end boards to begin with.

So, in the end, the TCO (Total Cost of Ownership) for a Raptor powered machine (even a 2x150 RAID Raptor setup) would still be significantly less than owning a similarly configured machine using a single 15K SCSI setup. And the Raptors will either perform just as well or they might outperform the 15K SCSI setup - and for the lower cost.
Click to expand...

Which is right, even though TCO still is a buzzword and purest garbage. (I have yet to see any 'TCO comparison' that's used valid criteria as opposed to marketing drivel.)
However, the fact is that your Raptors will be slower than even 10K Fujis on an LSI 20320RB in RAID1, forget even coming CLOSE to a pair of 15K Fuji MAX's on an Ultra320-2E (average sustained >120MB/sec, random seek <3ms what?)
Outperform? Keep dreaming, kid. Wrong wrong and yet more wrong. There isn't a chance in hell of a Raptor ever touching the performance of a Fuji MAX or Hitachi 15K147, ever.
 
I was wondering where AreEss had gone to! *dons flame retardant suit* :)
 
mwarps said:
I was wondering where AreEss had gone to! *dons flame retardant suit* :)
Click to expand...

Pimpslapping EMC repeatedly a few weeks ago for a "fault tolerant" director with "dynamic multipathing" that dropped a bunch of mission critical systems because neither claim was actually true. :D
Really, just been lurking and extremely busy. No real urge or time to post.
 
AreEss said:
Thanks for instantly disqualifying your arguments!



Bzzt, and thanks for playing. You already disqualified yourself, so you don't get to spout more shill from Tom's!
FACT: SCSI has and always will seek faster. Period. Faster RPMs, better algorithms, and superior command queuing.
FACT: SATA's only savior is it's huge buffers and the bell curve for such being around 20MB. A 10000RPM 16MB Hitachi FC-AL will happily rape a Raptor, then go for the money shot while the Raptor's still seeking.
FACT: SCSI doesn't ship on motherboards because your choices for integration are LSI (expensive) and Adaptec (garbage.) But you've NEVER had SCSI shipping on anything except high end boards to begin with.



Which is right, even though TCO still is a buzzword and purest garbage. (I have yet to see any 'TCO comparison' that's used valid criteria as opposed to marketing drivel.)
However, the fact is that your Raptors will be slower than even 10K Fujis on an LSI 20320RB in RAID1, forget even coming CLOSE to a pair of 15K Fuji MAX's on an Ultra320-2E (average sustained >120MB/sec, random seek <3ms what?)
Outperform? Keep dreaming, kid. Wrong wrong and yet more wrong. There isn't a chance in hell of a Raptor ever touching the performance of a Fuji MAX or Hitachi 15K147, ever.
Click to expand...
I think the real question that needs to be asked is "why"?

Why can't they have fast SATA drives? Why can't they have good algorithms? Sure SCSI provides much smarter controllers, command queuing which is now available in SATA and lower CPU utilization, but I don't see why they can't have fast seeking SATA/PATA drives..
 
Cost... SCSI drives are at the front the design curve for speed and reliability. Business customers purchase large quantities of these drives and replace them often while paying a price premium for the higher reliability and faster seek times.

If haven't priced out SCSI drives/hardware lately the cost per GB is anywhere from higher to considerably higher than SATA drives. The raptor is in many ways a nearline drive which adapts some technology taken from SCSI drives while a bit less (nevermind the 150Gb model with a damn window) and not requiring a separate controller card. When it comes down to it if you want some insane speed pickup a Fujitsu workstation 15k scsi drive and u320 controller card and watch the bits fly.
 
irrision said:
Cost... SCSI drives are at the front the design curve for speed and reliability. Business customers purchase large quantities of these drives and replace them often while paying a price premium for the higher reliability and faster seek times.

If haven't priced out SCSI drives/hardware lately the cost per GB is anywhere from higher to considerably higher than SATA drives. The raptor is in many ways a nearline drive which adapts some technology taken from SCSI drives while a bit less (nevermind the 150Gb model with a damn window) and not requiring a separate controller card. When it comes down to it if you want some insane speed pickup a Fujitsu workstation 15k scsi drive and u320 controller card and watch the bits fly.
Click to expand...

http://dailytech.com/article.aspx?newsid=6404
 
That test included a mix of all drive types including IDE. Going by the data google studied using just IDE drives on failure rates I would venture a guess to say that the statement about MTBF failure being over exaggerated in the Carnegie Melon study is probably false when applied to SCSI due them including IDE drives in their data set. Google found the failure rates on IDE drives to be roughly 12-15 times what their MTBF failure quoted, however which would fall in line with Carnegie Melons' findings skewed for IDE drives mixed with lower failure SCSI drives.

They did neglect to include data about duty cycle statements that IDE drive makers often add in the fine print which is really the main difference between 1,000,000 MTBF on a SCSI drive rated for a 24/7 constant read/write service vs and IDE drive rated for 1,000,000 MTBF running 8 hours a day for non-server duty (there have been articles written discussing this sleezy practice of manipulating MTBF numbers on IDE drives). That being said anyone that has run both IDE/SATA drives and SCSI drives in a data center environment will probably agree that they spend more time replacing IDE drives than SCSI drives in similar use servers.

As a final note I wouldn't say that I'm stating that SCSI drives don't also overstate their MTBF by some margin but SCSI wouldn't be around at all if it didn't have some benefits.
 
AreEss said:
Thanks for instantly disqualifying your arguments!



Bzzt, and thanks for playing. You already disqualified yourself, so you don't get to spout more shill from Tom's!
FACT: SCSI has and always will seek faster. Period. Faster RPMs, better algorithms, and superior command queuing.
FACT: SATA's only savior is it's huge buffers and the bell curve for such being around 20MB. A 10000RPM 16MB Hitachi FC-AL will happily rape a Raptor, then go for the money shot while the Raptor's still seeking.
FACT: SCSI doesn't ship on motherboards because your choices for integration are LSI (expensive) and Adaptec (garbage.) But you've NEVER had SCSI shipping on anything except high end boards to begin with.



Which is right, even though TCO still is a buzzword and purest garbage. (I have yet to see any 'TCO comparison' that's used valid criteria as opposed to marketing drivel.)
However, the fact is that your Raptors will be slower than even 10K Fujis on an LSI 20320RB in RAID1, forget even coming CLOSE to a pair of 15K Fuji MAX's on an Ultra320-2E (average sustained >120MB/sec, random seek <3ms what?)
Outperform? Keep dreaming, kid. Wrong wrong and yet more wrong. There isn't a chance in hell of a Raptor ever touching the performance of a Fuji MAX or Hitachi 15K147, ever.
Click to expand...

I have to agree. My old ass LSI MegaRAID U320-2 and a single 10,000rpm Fujitsu Ultra 320 SCSI drive are faster than a Raptor 150GB. I haven't done any RAID testing since I upgraded, but they aren't that much faster than the older 74GB drives would have been. I had actually benchmarked my 75GB Raptors against the above mentioned SCSI setup, and the single drive beat out the SATA RAID0 Raptor pair.

SCSI drives are superior, for all the reasons mentioned above. The other thing is that the onboard SATA controllers are crap compared to anything with a dedicated IO processor. Not that the difference would be worthwhile for most gamers and home users, but anyone who thinks that SATA and SCSI are in the same league, doesn't know what they are talking about.
 
You must log in or register to reply here.
Back
Top