SATA drive performance?

  • Thread starter Thread starter Colonel Blip
  • Start date Start date
C

Colonel Blip

Hello, All!

Planning my future purchases presents a question about SATA drives. All of
the info I read when these came out indicated you would pay more and get no
discernible improvement in performance. Now that things have matured a bit,
what is the experience here on SATA drive performance vs. straight ATA/133
drives?

Thanks.

Colonel Blip.
E-mail: (e-mail address removed)
 
Colonel said:
Hello, All!

Planning my future purchases presents a question about SATA drives.
All of
the info I read when these came out indicated you would pay more and
get no discernible improvement in performance. Now that things have
matured a bit,
what is the experience here on SATA drive performance vs. straight
ATA/133
drives?
Click to expand...


The same.

Performance is not about the interface, it's about the drive.

The first SATA drives were just the ATA drive with an ATA to SATA bridge.
This could limit speed a tad, but not noticeably. Now, many drives are
native SATA, but performance of a drive is mostly about the mechanics (seek,
spin speed etc), and partly about caching strategies (and cache size) etc.

I haven't looked into it for a while, but the fastast ATA drive used to be
the 74GB WD Raptor (I have the 36GB one). This drive is only available in
SATA, but the fact that it's the fastest is not due to the SATA interface,
but the 10krpm spin speed.

The cost is negligable, but the SATA wiring is much nicer, so go SATA if you
have a SATA interface on your motherboard. Quite when ATA will be phased
out, I wouldn't care to speculate, but that may be a consideration too.
Although adaptors between SATA and ATA are available (in both directions, I
think).

As a comparison though, I have the 36GB Raptor (GD), and the 250GB SATA
drive (JD). In terms of raw transfer speeds, they're pretty much the same.
However, if you attempt to delete a directory of several thousands of files,
or perform a text search in the same directory (on NTFS) - both operations
rely heavily on seek times, then the Raptor wins hands down, you don't even
need a benchmark program to tell the difference.

IMHO my setup is ideal, I have a fast drive for my OS and a large drive for
my media. If only I had the larger Raptor... :-p

Ben
 
Hello, Ben!
You wrote on Sun, 6 Feb 2005 13:01:53 -0000:

Thanks. That's a big help with my thinking at this stage.

Colonel Blip.
E-mail: (e-mail address removed)

BP> Colonel Blip wrote:
??>> Hello, All!
BP> The same.

BP> Performance is not about the interface, it's about the drive.

BP> The first SATA drives were just the ATA drive with an ATA to SATA
BP> bridge. This could limit speed a tad, but not noticeably. Now, many
BP> drives are native SATA, but performance of a drive is mostly about the
BP> mechanics (seek, spin speed etc), and partly about caching strategies
BP> (and cache size) etc.

BP> I haven't looked into it for a while, but the fastast ATA drive used to
BP> be the 74GB WD Raptor (I have the 36GB one). This drive is only
BP> available in SATA, but the fact that it's the fastest is not due to the
BP> SATA interface, but the 10krpm spin speed.

BP> The cost is negligable, but the SATA wiring is much nicer, so go SATA
BP> if you have a SATA interface on your motherboard. Quite when ATA will
BP> be phased out, I wouldn't care to speculate, but that may be a
BP> consideration too. Although adaptors between SATA and ATA are available
BP> (in both directions, I think).

BP> As a comparison though, I have the 36GB Raptor (GD), and the 250GB SATA
BP> drive (JD). In terms of raw transfer speeds, they're pretty much the
BP> same. However, if you attempt to delete a directory of several
BP> thousands of files, or perform a text search in the same directory (on
BP> NTFS) - both operations rely heavily on seek times, then the Raptor
BP> wins hands down, you don't even need a benchmark program to tell the
BP> difference.

BP> IMHO my setup is ideal, I have a fast drive for my OS and a large drive
BP> for my media. If only I had the larger Raptor... :-p

BP> Ben
 
Colonel said:
Hello, All!

Planning my future purchases presents a question about SATA drives. All of
the info I read when these came out indicated you would pay more and get no
discernible improvement in performance. Now that things have matured a bit,
what is the experience here on SATA drive performance vs. straight ATA/133
drives?
Click to expand...

It may be that I'm exceptionally dumb, but it's never been clear to me
why such a serial interfaced device (SATA) should be inherently faster
than a parallel interfaced device (IDE). I would have thought it would
be the other way around ultimately.
 
WoofWoof said:
It may be that I'm exceptionally dumb, but it's never been clear to me
why such a serial interfaced device (SATA) should be inherently faster
than a parallel interfaced device (IDE). I would have thought it would
be the other way around ultimately.
Click to expand...

At high speeds it turns out to be easier to use a serial interface,
because a parallel interface has problems like clock skew, where it
becomes too difficult to ensure that all the bits arrive at the other
end at the same time. Pretty much the same reasons why PCI Express is a
serial-type bus (albeit with multiple lanes). Aside from that the
cabling is a lot more convenient since fewer wires are required.

The other thing with parallel ATA is that the signalling is a
non-differential type with a high signalling voltage by current
standards (5 volts), which is difficult to increase in speed with a
reasonable cable length.
 
Robert Hancock said:
At high speeds it turns out to be easier to use a serial interface,
because a parallel interface has problems like clock skew, where it
becomes too difficult to ensure that all the bits arrive at the other
end at the same time. Pretty much the same reasons why PCI Express is a
serial-type bus (albeit with multiple lanes). Aside from that the
cabling is a lot more convenient since fewer wires are required.

The other thing with parallel ATA is that the signalling is a
non-differential type with a high signalling voltage by current
standards (5 volts), which is difficult to increase in speed with a
reasonable cable length.
Click to expand...

Another factor that is significent is the abililty to use different
controllers concurrently. If the source and destination is on the same
controller you can have a horrendous slowdown. The ability to have 3
separate controllers (4 counting firewire) can make a huge difference
especially if a swap file is involved. To get maximum performance you
should put each device on a separate controller. A drag and drop of a cd
from a master to a slave on the same controller is 10x to 100x slower than
to firewire drive or an SATA drive. At least that has been my observation
on the systems I own.
 
WoofWoof said:
It may be that I'm exceptionally dumb, but it's never been clear to me
why such a serial interfaced device (SATA) should be inherently faster
than a parallel interfaced device (IDE). I would have thought it would
be the other way around ultimately.
Click to expand...


Because these serial buses use a differential pair, the effect of
interference is greatly reduced. With parallel, you get cross-talk and and
other ill-effects of running wires in parallel.

Due to this you can clock it higher. A lot higher. High enough in fact
that you don't need to use more than 1 line.

And as already said, it's a lot easier to have a cable with 2 wires plus
plus maybe another couple overhead than 16, 32 or 64 parallel with overhead.

Ben
 
To the OP, I always used ATA/133 drives and found myself in need of
more storage. The only port left to install more HDs was the SATA. I
went ahead and installed a 200GB SATA with the same specs as my
ATA/133 drives. According to bench marks the SATA has a slower seek
time but a much faster average read time. Real world performance so
far also indicates much faster average write and read times. I'm very
satisfied with it so far. I also put my swap file on it to see if
that gives any performance increase and so far it seems about the same
but possibly a little lighter load on the ATA/133 drives where all the
apps are.
 
I have 2 WD SATA Raptors 10K RPM(36Gig each in Raid O Configuration), 1 60
Gig SATA 150, and 1 60 Gig ATA 133.

For comparison - here are the benchmarks (Note Access Time - Lower is
Better):

SATA Raptors
9.4 ms Access Time, 52.5 Meg/Sec Transfer, 7.6% CPU utilization

Single 60 Gig SATA 150
14.9 ms Access Time, 42.8 Meg/Sec Transfer, 5.9% CPU utilization

Single 60 Gig ATA 133
12.1 ms Access Time, 38.1 Meg/Sec Transfer, 5.4% CPU utilization

So as someone else in this thread pointed out - SATA has faster throughput,
but ATA has quicker access for such things as searches. Raid 0 of course is
the fastest of all (also the most CPU overhead). I also suspect that SATA
Raid is probably faster than ATA Raid.

Hope that helps :)
 
Tom said:
I have 2 WD SATA Raptors 10K RPM(36Gig each in Raid O Configuration), 1 60
Gig SATA 150, and 1 60 Gig ATA 133.

For comparison - here are the benchmarks (Note Access Time - Lower is
Better):

SATA Raptors
9.4 ms Access Time, 52.5 Meg/Sec Transfer, 7.6% CPU utilization

Single 60 Gig SATA 150
14.9 ms Access Time, 42.8 Meg/Sec Transfer, 5.9% CPU utilization

Single 60 Gig ATA 133
12.1 ms Access Time, 38.1 Meg/Sec Transfer, 5.4% CPU utilization

So as someone else in this thread pointed out - SATA has faster
throughput, but ATA has quicker access for such things as searches. Raid 0
of course is the fastest of all (also the most CPU overhead).
Click to expand...

RAID only uses CPU if you have a 'toy' software RAID controller. "REAL"
RAID controllers don't use any CPU time above normal I/O. These day's
there are so many surplus CPU cycles on most systems, it doesn't really
matter ... I'm just old-school and prefer real hardware RAID
controllers :)

FWIW, the gap between ATA133/SATA/SCSI RAID 0 performance drops as more
drives are added (assuming everything else stays the same) - eventually
you'll saturate the bus and there will be no performance gains in
throughput (except maybe in latency). Other RAID modes (1 and 5) are more
dependant on the host controller and/or CPU. RAID 1 especially will hit
the 'throughput wall' sooner than RAID 0 as it usually configured to do
parallel writes - making it particularly hard on PATA.

Cheers,

James
 
Hello, Tom!
You wrote on Mon, 07 Feb 2005 19:05:24 GMT:

Real DATA always helps!! <g>

Thanks.

Colonel Blip.
E-mail: (e-mail address removed)

TD> I have 2 WD SATA Raptors 10K RPM(36Gig each in Raid O Configuration), 1
TD> 60 Gig SATA 150, and 1 60 Gig ATA 133.

TD> For comparison - here are the benchmarks (Note Access Time - Lower is
TD> Better):

TD> SATA Raptors
TD> 9.4 ms Access Time, 52.5 Meg/Sec Transfer, 7.6% CPU utilization

TD> Single 60 Gig SATA 150
TD> 14.9 ms Access Time, 42.8 Meg/Sec Transfer, 5.9% CPU utilization

TD> Single 60 Gig ATA 133
TD> 12.1 ms Access Time, 38.1 Meg/Sec Transfer, 5.4% CPU utilization

TD> So as someone else in this thread pointed out - SATA has faster
TD> throughput, but ATA has quicker access for such things as searches.
TD> Raid 0 of course is the fastest of all (also the most CPU overhead). I
TD> also suspect that SATA Raid is probably faster than ATA Raid.

TD> Hope that helps :)

TD> ??>> Hello, All!
 
Colonel said:
Hello, Tom!
You wrote on Mon, 07 Feb 2005 19:05:24 GMT:

Real DATA always helps!! <g>
Click to expand...

Yes, but only when it has context, which that doesn't.

If he'd put actual drive model numbers down we may have had a chance of
comparison.

However, you have to throw out the RAID config, leaving one SATA against one
ATA, with those differences in seek time, you have to ask yourself why,
well... we don;t know. It could be the interface, but it could be rotation,
caching strategy, cache size... the list is pretty long.

Ben
 
It may be that I'm exceptionally dumb, but it's never been clear to me why
such a serial interfaced device (SATA) should be inherently faster than a
parallel interfaced device (IDE). I would have thought it would be the
other way around ultimately.
Click to expand...

The same thing occurred to me a while back and I found an excellent white
paper on the subject.

Serial interfaces are not inherently faster but that the component price per
bit is a lot less. Parallel leads have all sorts of problems with signal
noise/lag etc. These can be solved, which would indeed give a >serial speed,
but the cost is prohibitive.

They can use the same cost to improve the signal processing at each end of a
single serial interface.

Rob.
 
Robert said:
At high speeds it turns out to be easier to use a serial interface,
because a parallel interface has problems like clock skew, where it
becomes too difficult to ensure that all the bits arrive at the other
end at the same time. Pretty much the same reasons why PCI Express is a
serial-type bus (albeit with multiple lanes). Aside from that the
cabling is a lot more convenient since fewer wires are required.

The other thing with parallel ATA is that the signalling is a
non-differential type with a high signalling voltage by current
standards (5 volts), which is difficult to increase in speed with a
reasonable cable length.
Click to expand...

Interesting ... thanks for the explanation, Robert.
 
Back
Top