ATA and SATA are best suited for the lower end of the spectrum, while
SCSI and FC are high-end.  SATA still doesn't allow drives to
communicate without going through the controller.  SATA still doesn't
allow disconnecting a drive mid-spin and replacing it without
interruption of the system.  Hot-swap carriages cheat the system by
fooling the adapter into thinking a drive is there, but just busy, and
you could never replace a dead one with a drive of differing capacity.
Here's a real example of hot swap:  if I wanted to add several terabytes
to my SCSI/160 RAID10 stack here, I can do so, replacing one drive at a
time with larger drives.  I don't have to shut down the array, and the
attached servers don't even feel a slowdown in performance.  With all
ATA/SATA systems, you must shut down and rebuild the array.  On-the-fly
scalability is critical in the high-end market.

No, I'm really not selling ATA or SATA short, but most ATA/133 drives
(the hardware itself) were never designed to drive throughput anywhere
near 133, and most SATA drives will never drive throughput at anywhere
near 150.  Your example of 450 MHz with 8 drives connected is a good
example-- even using ATA/100, shouldn't an eight channel RAID 5 be able
to handle a sustained throughput of 700?  Well, no, because it's ATA.

I'm sure SATA will hit 300 and 600 just like processors went from 4.77
MHz to 4.77 GHz.  But when all of us were dazzled by machines that
topped 100 MHz, the big machines were already playing at over 4 GHz.
Now we are in awe when our desktops hit 4 GHz, and the big machines are
dealing in teraflops.  To propose a SATA drive for one of these big
machines would be just as ridiculous as proposing a fiber channel RAID10
stack for a workstation.  99.999% of the processor time would be used in
waiting for the drives.  As desktop systems get faster and faster, so
will the standards.

I personally believe the biggest advantage with SATA is its support for
extremely large drives.  Maxtor is coming out with a Terabye drive in
the next few years, and it will not work with ATA.

As for standards in general, it's always important to remember that
standards have one purpose: "sell lots of hardware".  This is
accomplished by everyone agreeing on how to interconnect.  Don't be
fooled into thinking that they create standards for any other reason,
otherwise everything would be connected with FireWire, and we'd all be
using Apple servers. :)

The manufactures supporting SATA will never allow it to compete against
their high-margin products, SCSI and FC, so purposefully it will never
be quite as fast, quite as robust, or quite as capable.  You will always
be able to find premium (i.e. $15-$25 per gig) SCSI and FC products that
are better than anything made for ATA or SATA.  It's all about who is
willing to spend the most money.  Profit margins in the ATA and SATA
market are extremely tight, while SCSI and FC are very generous.

As for choosing a RAID type, it always depends on the usual argument of
budget vs. performance requirements vs. risk tolerance.  RAID5 is great
if you are okay with the risk that if two drives fail at the same time,
it's dead.  With very big arrays, your risk of dual failure is a lot
higher, especially if you bought all of the drives at the same time and
from the same place.  RAID50 is better than RAID5.  I'm still of the
opinion that RAID10 is the best for reliability and performance if you
have the budget, especially if they are SCSI so the mirrors are updated
across the cable and not through the adapter.

As for threads and whatnot, that also depends on your controller, its
cache size, the machine it's attached to, the speed of the drives, and a
lot of things.

 

> -----Original Message-----
> From: [EMAIL PROTECTED] 
> [mailto:[EMAIL PROTECTED] On Behalf Of Matt
> Sent: Thursday, March 25, 2004 10:08 AM
> To: [EMAIL PROTECTED]
> Subject: Re: [Declude.JunkMail] Raid Controller
> 
> You are selling serial ATA short.  Also, I'm not sure if you 
> are mistaking ATA with serial ATA in your reply.
> 
> It does turn out that there is some logical speculation that 
> SCSI drive manufacturers are treating SATA as their economy 
> server/workstation class, however Western Digital's 10K drive 
> has no in house SCSI alternative.  Tom's did a comparison of 
> that to the Seagate here:
> 
>     Smart Hard Drives: Seagate Barracuda 7200.7 and Western 
> Digital WD740 Raptor
>     http://www.tomshardware.com/storage/20040123/index.html
> 
> I am definitely going the Western Digital route based on what 
> I saw there.
> 
> Regarding SATA RAID cards, there are two things happening 
> here.  First, 3wave and LSI are both working on full featured 
> versions for SATA, and they are starting to support native 
> command queuing which apparently can speed performance by 
> 1/3, and instead of bridging ATA to SATA on the hard drives, 
> these are now also being made natively now as well.  Here's a 
> press release concerning LSI's upcoming offering:
> 
>     LSI Logic launches industry's first PCI-X enabled 
> hardware based MegaRAID SATA 300-8X solution
>     http://www.lsilogic.com/news/product_news/2004_02_17a.html
> 
> I don't think you can find any fault with that.  Concerning 
> the bus bottleneck, the PCI-X upgrade will take care of that 
> on a capable system.
> 
> I did some more research though and found the following 
> review of a new approach to RAID from some former Adaptec 
> employees (now owned by Broadcom).  It's a company known as 
> RAIDcore and their performance is at least on par with 
> Adaptec SCSI in the benchmarks on a RAID 0 installation 
> according to Tom's Hardware
> 
>     RAIDCore Unleashes SATA to Take Out SCSI
>     http://www.tomshardware.com/storage/20031114/index.html
> 
> Better yet, the card has 8 independent channels and you can 
> span across controllers among other things.  I plan on buying 
> one of these and doing RAID 50 which will give me redundancy 
> plus speed without having to dedicate a disk to a specific 
> drive.  This is generally cost prohibitive with SCSI.  
> RAIDCore claims performance of 450 MB/s sustained reads, and 
> 230 MB/s sustained writes using an 8 drive RAID 50 setup.  
> The total cost for the drives plus the card would run me 
> $1,200 (no hot swap, though that is available, even Intel is 
> coming out with hot swap SATA drive carriages this quarter).  
> A comparable setup with SCSI RAID 50 would run 3 times the 
> price and might not out perform.
> 
>     http://www.raidcore.net/RC4000DataSheet_2.pdf
> 
> The only issue that I see with this is the company is young 
> and this is their first product (though they are backed by 
> Broadcom now), but it looks real hot and you can get an 8 
> channel card for under $400.
> 
> If you are wondering about the effect of RAID 50 over just 
> plain RAID 5, here's a nice start:
> 
>     http://cdfcaf.fnal.gov/doc/cdfnote_5962/node15.html
> 
> This shows that while performance decreases with RAID 5 as 
> the number of threads increases, RAID 50 maintains it's 
> throughput until around 40-60 read threads before it drops 
> off.  Naturally this would be somewhat unique to each card, 
> but it makes plenty of sense.
> 
> I believe that SCSI is just a physical interface/transport 
> layer protocol if I'm not mistaken, and all that makes SCSI 
> special is what's connected to it at either end.  SATA is 
> more capable, and SATA II which does 150 MB/s will be 
> replaced by 300 MB/s versions, 600 MB/s versions, and on, but 
> for now, there is no need for even 100 MB/s on one channel in 
> this configuration so that doesn't matter.  It's foolish to 
> think that SATA won't take over the market as soon as they 
> start connecting the good stuff to SATA wires, and it looks 
> like they are starting to do that now.
> 
[AUTOMATED NOTE: Your mail server [63.147.33.8] is missing a reverse DNS entry. All 
Internet hosts are required to have a reverse DNS entry. The missing reverse DNS entry 
will cause your mail to be treated as spam on some servers, such as AOL.]

---
[This E-mail was scanned for viruses by Declude Virus (http://www.declude.com)]

---
This E-mail came from the Declude.JunkMail mailing list.  To
unsubscribe, just send an E-mail to [EMAIL PROTECTED], and
type "unsubscribe Declude.JunkMail".  The archives can be found
at http://www.mail-archive.com.

Reply via email to