I'm probably buying my new box in about a month, and I'm definitely going for a SATA drive with NCQ. I'm a bit curious about this - it's been said that a 7200rpm drive can approach 10.000rpm performance because of the command re-ordering. I can see why NCQ improves performance, but 7200->10.000rpm does sound a bit optimistic.

So... does anyone here have a SATA drive with NCQ, and how is performance? :)
Posted on 2005-01-07 11:28:21 by f0dder
I have heard that they did a lot of mistakes in SATA specifications an implementation.

One of the main is using plain not shileded wires for data transfer. This causes data to be lost sometimes :P

10.000 rpm might improve speed a little (aka you have to wait less until the next sector comes under the head.

But higher rotational speed also means faster mechanical failure and uzure, also bigger noise levels.

Of course this is acceptable for servers because of RAID and you are keeping them ON all the time and in a sepparated room. Also on a server you assume that you will update the HDD next year anyway.

SATA is supposed to be only marginally faster that normal PATA. Form my tests in practice it is not.

The whole idea of using serial instead of paralel cable is designed to reduce the costs for the manufacturer ;)

One must be very stupid to assume that serial data transmision is faster than paralel. And i will smile a lot when over the next years somebody will "invent" the "new" and "faster" paralelSATA ;)

I did the thest on a SATA on a job i have had a few months ago... maybe things have improved since.

The only nice thing i have noticed about SATA were the less cables inside the case... like i would really care.

So if I would have the option, i would really wait a little until SATA technology becomes more mature and they do something about the data loss and iterference on the non shielded cables
Posted on 2005-01-07 22:46:49 by BogdanOntanu
I was talking about "Native Command Queuing" which lets the drive hardware re-order read/write requests, in order to decrease the amount of head movement necessary. Of course an OS will already sort it's requests if it's worth anything, but the OS doesn't know about the internal drive layout (number of plates, heads, etc.).

So I wasn't talking about going from 7200->10k RPM, but that it's said the NCQ feature should give a similar improvement in seek time, plus less mechanical stress.

NCQ is a pretty new feature btw, and requires both chipset and drive support, so the box you tested might not have had it. It's also worth to notice that most SATA drives, at least initially, were just PATA drives with a SATA interface.

Non-shielded wires, ho humm - that does sound a bit silly. But I'm pretty certain that the PATA and SATA protocols include checksums and lost data will cause re-transfer... so while it will slow down, it won't give corruption. I do care about having smaller cables, it improves airflow in the cabinet.

SATA is supposed to be only marginally faster that normal PATA. Form my tests in practice it is not.

The drives haven't changed all that much, you rarely reach the maximal throughput on ATA133... what does matter about SATA is when you start building large RAID arrays :), or if the NCQ feature really does give better seek time.

One must be very stupid to assume that serial data transmision is faster than paralel.

Why, if they can drive the serial transfer speed much higher than the parallel speed? USB is a serial bus and it can theorically achieve 480 mbit/s. PCI express is also a serial bus, and is faster than regular PCI, but can be multiplexed (or "parallel-serialized" ;)) for graphics cards...

I'm not an engineer, but I guess the use of serial transmission gives less EMI? Then again, with higher data rates... ho humm.

And hey, if it's cheaper for the manufacturers, perhaps it'll be a bit cheaper for us too (dreaming).
Posted on 2005-01-07 23:03:09 by f0dder
I also read the main advtange of STA over PATA is that in PATA speeds cant be pushed as far because of the crosstalk that will be experienced by the lines. SATA eliminates this problem.
Posted on 2005-01-08 14:20:39 by x86asm
SATA drives also are specified as to their connector positions (ATA drive connectors "wander" from model to model, let alone mfr to mfr), and are "staged" (ground, then power, then signal), allowing correct hot swapping (also enabled by each drive being an ATA "master", with no shared signal connections).

My speed tests have shown them to be the same as current generation PATA drives. If the requested sector(s) are cached, the limit is the SATA interface speed. If not cached, the limit is the seek and rotational latency. On long reads, the limit is the platter to head read speed, which varies by about 2:1 from the outside to the inside tracks. All are close or identical between PATA and current generation SATA.
Posted on 2005-01-23 02:18:14 by steverose