SCSI vs. SATA What's better?

what's better.. it depends.. if cost is no issue then I'd say SCSI. but if cost is then Sata..

Current scsi drives are U320- which is 320MBps per second versus SATA's rate of 150-- BUT the SATA standard is scalable to 600

According to the SATA Working Group, SATA defines a roadmap that starts at 1.5 gigabits per second (equivalent to a data rate of 150MB/s) and migrates to 3.0 gigabits per second (roughly 300 MB/s), then to 6.0 gigabits per second (600 MB/s) -- six times faster than the current ATA standard

SATA is meant to replace the current PATA tech but not scsi.
 
depends on what you're doing. scsi will let you hook more devices to a channel and the drives are, generally speaking, more robust and faster in a workstation or server environment (the new 74GB SATA raptor is an exception to the workstation performance rule). scsi is also more expensive

if you just want a fast hard disk for a desktop, the raptors are great. if you're willing to pay more for a drive that might be more reliable, go with scsi.

i personally like scsi drives and would most likely be willing to spend the ~300 or so for a 36 gig maxtor 15k drive (or whatever is out) the next time i upgrade
 
SCSI is better if you need the best performance and the highest reliability.

SATA is better if you need fairly good performance and decent reliability, but need something low-cost.

SCSI controllers generally take a lot of the processing overhead off of the host system processor. Additionally, more devices can be connected to a SCSI controller than your typical SATA controller. (maybe 2 or 4 on SATA compared with 15 or 30 on SCSI)

SCSI drives are also targeted more toward enterprise servers and workstations where reliability is a big concern. SATA drives are generally built to be cheap, which typicaly ends up decreasing their reliability.

If you want to see what is probably the only drawback of SCSI, take a look at the prices of similar capacity drives in both SATA and SCSI. You'll see that typically SATA drives are not even half as much as SCSI drives.
 
ahhhh, very good information! thanks guys. :D Yeah for sure, it sounds like SCSI would be the way I want to go (obviously), but cost is a concern. I really like the idea of the SCSI controller taking a load off of the processor though, SATA doesn't do that? If I go with SATA for now, I want to be able to have the option to hook up 4 drives though. That way I can RAID two of them and put data storage drives on the other two cables. I'm looking at the ASUS P4C800-E Deluxe and the ABIT IC7-Max 3 mobos.

This is what newegg says about the P4C800-E -
-Onboard SATA/RAID: ICH5R, 2x SATA, RAID 0/1; Promise 20378, 2x SATA, RAID 0/1/0+1/Multiple RAID

And This is what they say about the IC7-Max 3 -
-Onbaord SATA/RAID: 2x SATA RAID 0/1 via South Bridge; 4x SATA RAID 0/1/0+1 via Silicon Image Chip

Thing is I don't understand what Promise 20378, South Bridge, and Silicon Image Chip all are. Can anybody help explain this for me?

Also, could I hook up 4 SATA Drives to either of these boards? Or just the Abit?

Also, would it even make any difference to have my data drives be SATA? Meaning if I put twin raptors in RAID0 in my SATA slots and I put a 200gig ATA data drive on IDE, would the transfer rates be the same or any different compared to if I used a SATA data drive on the SATA cables? Of course the difference would be going from ATA100 or 133 to 150 which is better, but I've heard something about most drives not being able to transfer that fast anyway.

Thanks!
 
The South Bridge is the secondary chipset chip that typically serves as an ISA/PCI bridge and hard drive controller among other things.
With the Asus board the Intel ICH5 South Bridge (part of the chipset) has two SATA ports (for two drives), and two on the additional Promise controller. Both motherboards have this south bridge.

Promise and Silicon Image are companies that make dedicated SATA controllers (that are not implemented in the south bridge).

The Silicon Image in this case is inherently better as a controller, so I would suggest it over the hybrid translated (IDE to SATA) Promise.
 
Serial Attached SCSI

When it becomes available....:D

They use the same cabling and SAS controllers can handle regular SATA drives since both SAS and SATA are compatible on the physical layer. You get all the benefits of SCSI and then some.
 
Id say it really depends on the application
its fairly rare for a desktop or workstation to need alot of performance drive space, but often needs a fairly large amount of data storage, so likely a combination (if you can afford it) would be best

but deciding if you really need SCSI performance levels
would be determined by your aps

the interface (U320 @ 320MB/s vs SATA 150MB/s) is misleading in that no single HDD in either class can actually fill that bandwidth
(youd need multiple drives in a RAID array to do that) and that those are just burst speeds (in addition a U320 card on a typical PCI 32bit 33MHz bus\slot would be bottlenecked to 133MB/s anyway)

review As The Disks Spins @ Lost Circuits
for a comprehensive explaination of HDD performance metrics

also review the RAID guides, ROMB (RAID on Mother Board) is really sort of a form of software RAID the CPU doing most of the work, this is true even in many ATA RAID cards at simple RAID levels (0, 1, 0+1) but is considerably different for SCSI RAID implementations


now to sit back and wait for UICompE02's SAS post :p
 
so for the purpose of rendering would having a RAID controller card be more advisable than simply using the onboard RAID since rendering is generally cpu intensive and if ROMB is using up cpu then the assumption from your post would be that onboard RAID would adversely affect rendering performance.

hmmmm...that sure was a long sentence, must have missed a few periods, commas and conjunctions.

Oooh, conjunction junction what's your function. lol.
:p
Sorry, couldn't help it.
 
yup, a SCSI RAID card implementation with its own dedicated processor or an external SCSI array would be ideal
and would certainly have less CPU impact than an onboard ATA RAID controller, the specific implementations vary however even in SCSI, so thats an area to research when considering a solution

RAID Chapter I: The Lesser Levels

Different “Categories” of RAID

In general, one needs to distinguish between three different categories of RAID, namely:

1. Standalone RAID solutions for mass storage in the backplane of servers (fiber channel-attached) or Network-Attached Storage (NAS) connected via Firewire or Gigabit Ethernet.

2. RAID functionality via separate Host Bus Adapter (HBA) cards using PCI, PCI-X (64bit / 66 MHz) or PCI-Express (3GIO) interface.

3. RAID controller integrated on the mainboard level. Often called RAID-lite because of limited functionality, except for dedicated Server boards. RAID on Mainboard (ROMB).

Software vs. Hardware RAID

Software RAID


In general, both Striping and Mirroring, that is Level 0 or Level 1 RAID, respectively, can be achieved in software alone as long as the operating system supports it. This type of RAID is also called zero-channel RAID (because no dedicated hardware channels are involved)

Hardware RAID

Whether any given RAID solution is Hardware or Software often boils down to a matter of semantics. That is, by definition, any dedicated hardware RAID channels will qualify the array as Hardware RAID. However, often enough another distinction is made based on the presence or absence of a dedicated XOR processor. That is, any hardware RAID controller without the additional controller will often be referred to as software RAID, in other words, there is a grayzone. We will cover the significance of the XOR operations a bit later, for now, let’s start with the “lesser” levels of RAID, that is Level 0, 1, 0+1 and JBOD.

another point of consideration is the PCI bus bandwidth which is touched on in both these posts and boils down to this
there are few single user applications that can actually saturate a higher bandwidth bus even with a RAID array that could, it typically takes a multiuser environment, but some aps (like video editing) easily exceed the typical 32bit 33MHz bandwidth when used with a RAID 0 array (especially if the stripe width, that is the number of HDDs exceeds 2) in fact saturating the 32\33 PCI is pretty easy with a RAID 0 array even with 2 drives
--------------single HDD ATA 50MB/s SCSI 15k 80MB/s STR (best sustained transfer rate)
PCI 32bit 33MHz 133MB/s
-------------dual HDDs in a RAID 0
-------------U160 SCSI 160MB/s HBA or RAID controller
PCI 64bit 32MHz 266MB/s (or 32bit 66MHz)
-------------U320 320MB/s HBA or RAID controller
PCI 64bit 66MHz 532MB/s
PCI 64bit 133MHz 1GB/s (PCI-X)

so depending on the stripe width, but also on the access the ap is trying to read\write and the HBA own internal limitations you can easily use bandwidth into the 64bit 66MHz zone, but probably not saturate it

however with rendering your primary access and bandwidth useage will be virtual memory (pagefile) and regardless of the stripe width and bus speed its still much much slower than the system memory
so increasing your RAM (and the OS's ability to address it) pays much bigger dividends
Breaking Through the 2 GB Memory Ceiling @ workstationplanet
 
Originally posted by shadow74205
Serial Attached SCSI

When it becomes available....:D

They use the same cabling and SAS controllers can handle regular SATA drives since both SAS and SATA are compatible on the physical layer. You get all the benefits of SCSI and then some.

My suggestion as well :)

SAS controllers, drives, and expanders should be available sometime near the end of the third quarter this year.

The goal of SAS was to provide a solution that was faster than U320 SCSI while not increasing the cost. Now, this doesn't mean that controllers and drives will be cheaper, but Serial SCSI cables are much cheaper than Parallel SCSI cables. (in fact, as shadow mentioned, they are the exact same cables that SATA uses, plus you don't need terminators any more!)

I guess the feature I'm most excited about is the SATA compatibility. Imagine it this way, you can have a fast, somewhat smaller, very reliable SAS drive for your OS and important applications and data and put your huge media files on a giant inexpensive SATA drive -- all on the same controller! (okay, admittedly, I get excited easily...)

First generation SAS controllers will have four ports to start with, then eight port controllers will be available shortly thereafter. But you can greatly increase the number of devices attached to the controller with devices called expanders that will typically multiply a set of four ports to twelve, for example. (and on each of those ports could be either a SAS or SATA drive)

now to sit back and wait for UICompE02's SAS post

There it is :) How'd you know?
 
Originally posted by Ice Czar
n general, both Striping and Mirroring, that is Level 0 or Level 1 RAID, respectively, can be achieved in software alone as long as the operating system supports it. This type of RAID is also called zero-channel RAID (because no dedicated hardware channels are involved)

I always questioned this paragraph from that website. The reason being that Zero Channel RAID is typically controlled by hardware, not software. For example, the ZCR solutions I have seen all involved a standard non-RAID SCSI controller (with a couple of pins hijacked for ZCR usage, typically the PCI TDI and GNT signals) that was controlled externally by another IO controller (Almost always an Intel IO Processor of some sort)

Typically ZCR controllers are sold as an add-on for certain systems that have special SCSI controllers on-board or a ZCR PCI slot.

Okay, enough off-topicness for one day...
 
Originally posted by UICompE02
I always questioned this paragraph from that website. The reason being that Zero Channel RAID is typically controlled by hardware, not software. For example, the ZCR solutions I have seen all involved a standard non-RAID SCSI controller (with a couple of pins hijacked for ZCR usage, typically the PCI TDI and GNT signals) that was controlled externally by another IO controller (Almost always an Intel IO Processor of some sort)

Typically ZCR controllers are sold as an add-on for certain systems that have special SCSI controllers on-board or a ZCR PCI slot.

Okay, enough off-topicness for one day...

I figure Micheal was addressing desktop use of that term
as opposed to the server boards,
except for dedicated Server boards.
which he broke out earlier, and of course the series current focus on ATA
but Ive found him very open to input and suggestions, however if he finds out your working for LSI and developing the SAS implementation he might suck your brains out with a straw :p

I'll PM you his email and an introduction ;)


There it is ;) How'd you know?

Im psychic :p

very informative ;)
 
Originally posted by Ice Czar
I figure Micheal was addressing desktop use of that term
as opposed to the server boards,

Well, that explains that.

Similar to how everyone I've worked with uses JBOD to refer to any generic multi-disk storage enclosure rather than the way it's used in the rest of the world to mean a storage volume that spans several disks with no special striping or redundancy -- It all depends on who you're talking to and what you're talking about, I guess.
 
I use SCSI for my system drives since they are faster and seem to be a lot more reliable.
I use IDE for storage since the price/gig can't be beat with SCSI.
I've been running SCSI and IDE in my systems for years with no problems. I actually started with SCSI since my Amiga3000 came with SCSI and there wasn't an IDE controller available for it back in 92.
 
Originally posted by UICompE02
It all depends on who you're talking to and what you're talking about, I guess.

Think thats a really good analogy, and Ive seen JBOD used like that too.
"Err let me get this straight Im running a JBOD on a JBOD?" :p

In this case I figure Micheal was using ZCR simply to describe RAID using onboard channels
but the more I think about it the more I agree with you (and he might too)

WS6, ask UICompE02 to relate his personal SCSI reliability observations ;)
 
Originally posted by Ice Czar

WS6, ask UICompE02 to relate his personal SCSI reliability observations ;)

No need to ask even, here's a snippet I posted in a similar thread a few months ago:

Sure, this evidence is somewhat anecdotal, but we do have a significant number of disk drives of all types in our hardware engineering labs.

Here's a specific example, I was working on some stress tests for a U320 SCSI to Serial ATA enclosure that was being developed by a customer of ours. We had 4 of these enclosures, each running with 14 WD Caviar SE SATA 250GB drives in them (actual production models and firmware). (a total of 56 SATA drives)

By the time we had all of our setup and firmware issues (enclosure FW, not drives) solved and ready to go, three of the drives were already dead. Well, this didn't bode well for the future of our stress testing. We removed those drives from the enclosures and continued testing. Granted these are hard 24/7 stress tests, that SATA wasn't really meant to handle... After that point, we did about two weeks of stress testing and lost 3 more drives over that period.

That would put our SATA failure rate for this particular test at more than 10% for a two week period. I think once we weeded out the weak drives, though, everything was pretty much stable afterwards.

Now, contrast this with my latest U320 SCSI tests. I have personally seen only one of these drives fail in my life. And keep in mind we are running, by far, more SCSI stress tests than SATA tests. (and, in contrast to the SATA drives, many of these are pre-production models)

We continue running tests with SATA drives as we are testing the STP (SATA Tunneling Protocol) capability of our SAS controllers and from what we are observing, the 10% failure rate for new SATA drives seems to be holding up. Consensus of the engineers in our lab seem to indicate that they are, and I quote, "crap."

This is why I began even more regularly backing up my data on my home computer...

Of course in a RAID configuration (not RAID 0, of course), it's not as risky. But you have to ask yourself how much these drives are going to be used. It seems to me that if they are going to be turned on 24/7 and used fairly heavily, after accounting for the SATA drive failure rate, it may be cheaper to get SCSI drives in the first place rather than keep spending the money and time to replace SATA drives that may not be as reliable.

I've already heard twice this week other engineers stating that the SATA drives we are working with are "crap." And it's only Tuesday...
 
wow, very good info! thanks guys! i'm learning a lot.

also review the RAID guides, ROMB (RAID on Mother Board) is really sort of a form of software RAID the CPU doing most of the work, this is true even in many ATA RAID cards at simple RAID levels (0, 1, 0+1) but is considerably different for SCSI RAID implementations
so for the purpose of rendering would having a RAID controller card be more advisable than simply using the onboard RAID since rendering is generally cpu intensive and if ROMB is using up cpu then the assumption from your post would be that onboard RAID would adversely affect rendering performance.
yup, a SCSI RAID card implementation with its own dedicated processor or an external SCSI array would be ideal
and would certainly have less CPU impact than an onboard ATA RAID controller, the specific implementations vary however even in SCSI, so thats an area to research when considering a solution

Would this be true with SATA RAID too? Would it make any difference in the board you get? Like if I get the IC7-Max 3, would it still be better to get a separate SATA RAID controller card than use what's on that board? I guess I'm thinking that the Silicon Image Chip would be comparable to just having a separate card, am I wrong? Would that chip take over the controlling so my processor doesn't have to do it?
 
about SATA being crap, I'm assuming that is in comparison to SCSI drives. How about comparison to good old IDE? SATA should be an viable alternative to IDE (I guess with a few pitfalls). am I wrong?
 
highend RAID Cards typically have their own CPU
for instance my IDE RAID card (Promise SX6000) has a RISC processor and a 128MB cache for reads and writes
mostly that is employed in the higher RAID levels 5 & 3
( the cache is needed while the parity calculations are underway), however while simple striping (0) or mirroring (1) or a nested array of both (0+1) dont require parity calculation, but they still impact the CPU (as does the onboard IDE)
so offloading those to a dedicated CPU has advantages, but its far more noticeable with the higher RAID levels

SATA and PATA (IDE) is just the interface and really has to deal with RAID in the same way

the SCSI part of this question I'll leave to UICompE02
but I do know they also typically have dedicated processors as well (or at least many)
 
Originally posted by mayakindaguy
about SATA being crap, I'm assuming that is in comparison to SCSI drives. How about comparison to good old IDE? SATA should be an viable alternative to IDE (I guess with a few pitfalls). am I wrong?

well keep in mind that UICompE02 is
the Grand Inquisitor in the LSI Logic HDD & RAID Card Torture Chamber,
he tries to break stuff :p

Its not suprising the SATA drives dont hold up at the same rate as SCSI drives, they are really designed for different missions
Are SCSI drives more reliable than ATA drives?

again PATA (IDE) and SATA are pretty much the same thing
as far as the physical HDD itself goes
 
Originally posted by Ice Czar
well keep in mind that UICompE02 is
the Grand Inquisitor in the LSI Logic HDD & RAID Card Torture Chamber,
he tries to break stuff :p

Hey, who told you my official title at work? :cool:

Yes, when we are stress testing our controllers we are running the hard drives attached to them at the fastest they'll go for 24/7 for several weeks.

SATA and IDE are not really intended for this type of use. Most manufacturers even state they are designed for 5x8 use, not 24x7.

That being said, the SATA drives we deal with seem to have a not insignificant infant mortality rate (meaning they die pretty early on in the tests), but it seems the drives that make it past the initial burn in will last for a while longer.

There are some other reasons SCSI and SAS are more reliable than SATA, mostly regarding error reporting and handling in the command set (SATA/IDE basically have almost none (save for the fairly unreliable SMART), whereas SCSI/SAS devices will give you very detailed information on why they failed)

There are other benefits to the SCSI/SAS protocols over ATA... Shall I write a FAQ, Ice Czar? :)
 
Originally posted by Ice Czar

the SCSI part of this question I'll leave to UICompE02
but I do know they also typically have dedicated processors as well (or at least many)

Almost all the current generation (Ultra320) SCSI RAID controllers I've seen have their own processor that handles the DMA transactions and parity generation.

There are a few SCSI RAID cards out there that depend on the host processor to handle a lot of the RAID overhead, but these are typically only capable of RAID 0 and 1. One example would be Adaptec's HostRAID that is offered on their U320 controllers.
 
Got both and SCSI everytime although SATA is realy nice as well. Maybe SATA2 will be better.
 
Originally posted by UICompE02
There are other benefits to the SCSI/SAS protocols over ATA... Shall I write a FAQ, Ice Czar? :)

Yes by all means :D :D :D

Ive been working on hacking down our Sticky thread mess
(think we have more sticky threads than any other forum :p )
Planning on rolling many of them into a comprehensive FAQ
along the lines of what Bob Sultan and RichardParry have done with the Networking FAQ

others to be rewritten and new topics to be addressed
(and a complete revamp of the Advanced HDD Issues LinkFarm)

Id be real interested in as much SCSI FAQ material as you would care to write, from the highly technical to the n00b level
 
Believe it or not, but I have a 640meg SCSI Micropolis that I got used from a friend back in 93 or so and it still works to this day, although I don't use it since it doesn't hold much. I had it in my server back when I had a 24/7 Dial up connection and runing an FTP server. I guess you could call my server a router as well since I used it along with WinGate to share the internet connection with 2 other machines.
 
SATA and IDE are not really intended for this type of use. Most manufacturers even state they are designed for 5x8 use, not 24x7.

What's really interesting to me is when EIDE/SATA drives are explicitly specified for 24 hour operation, like the MaxLine Plus II series.

Does this mean anything at all about the drive models actual MTBF?
And how will Fibre attached ATA fit into all of this?
 
kinda on topic, I was playing around with a pair of wd800jb
(not raid) conected thru these adapters to my
ICH5R sata headers. to make a long story short, I was getting slightly faster transfer rates than a pair of u160(1 10k 1 15k) drives on
one channel of an adaptec 29160 controler card (in a slow pci slot)
Fire away:D "gets ready for flames"
 
Originally posted by leukotriene
And how will Fibre attached ATA fit into all of this?

FATA drives will fit into what is called "near line storage." That is, it is intended for large storage that isn't accessed as often but can go into a Fibre Channel topology. It's sort of like how IDE and SCSI or SATA and SAS go together, but using a Fibre channel interface.
 
Back
Top