USB 2.0 disappointment

[LanEVENT]

I got my HD Enclosure that uses FireWire or USB 2.0 ? I taken these 2 pics from Sandra of performance USB 2.0 - FireWire (I also used HDTach, and WinBench 99 and got the exact same results)

For a point of reference:
FireWire 400Mbit/sec = 50Meg/Sec (theoretical max)
FireWire 480Mbit/sec = 60Meg/Sec (theoretical max)

However, it seams that USB2.0 gives almost exactly half the performance of FireWire under HD Access patterns. What?s more, with the first drive I tested (an old Caviar 2600 2.0gig drive) I noticed USB2.0 had terrible lag and writing data (It would pulse about ever 10 seconds around 5 megs of data, where under the same conditions the FireWire would have a constant stream and be done in about 10 seconds) I switched that with a 30 gig IBM 7200 RPM and this particular problem went away.

I have 3 separate brands of USB2 Cards, the one that came with the drive (no name), and Adaptec, and SIIG. However they all have the same chip (NECD720100AGM as it is the only one to pass WHQL - Adaptec has NECD720100AS1, which appears to be the same chip in a smaller package - same drivers install)

I am hoping that it?s the USB2 controller on the actual drive enclosure that may be causing the problem, however it handles FireWire wonderfully. But half the speed of something that is supposed to be inferior is bothersome.

Any thoughts or Info on what may be the issue?

 

[LethalWolfe]

Originally posted by: LanEVENT


For a point of reference:
FireWire 400Mbit/sec = 50Meg/Sec (theoretical max)
FireWire 480Mbit/sec = 60Meg/Sec (theoretical max)

Errr... typo?


IIRC, USB has more overhead than Firewire and that degrades performace. There might be something defective w/yer enclosure I don't know. But from what little I've read USB 2.0 ain't all it's cracked up to be. Personally, I'll stick w/Firewire and the upcoming Firewire 2 (which is going to double the bandwidth) and USB 1 for my low bandwidth needs (KB & M).


Lethal

 

[GoodRevrnd]

Doesn't USB offload much work onto the cpu while firewire will have a dedicated controller? I don't know that that could make such a big difference, esp. on a dedicated task, but is it a possibility?

 

[Adul]

firewire ha a dedicated controller and usb relies a lot on the cpu, USB 2.0 is nice, but ti cant hold a candle to firewire.

 

[LanEVENT]

Well, I have a 2.0 Ghz P4, and the CPU use was minimal. The NEC chip is extremely large, one would think it would be handling USB instead of just off loading it to the CPU.

Oh well

 

[Lord Evermore]

The size of the chip doesn't determine anything, as that is affected by the process size used to create it as much as how many transistors it has. Also, the NEC chip in the enclosure is only used to convert the IDE signal from the drive into a USB or Firewire signal, and vice versa. It doesn't do anything related to the actual data transfer from the system to the firewire controller.

It is inherent in the USB specification that the CPU does a lot of the processing of data; all the transactions have to go through the CPU, whereas with Firewire, you can have one device like a DMA hard drive directly communicating with the firewire controller through the main chipset, without the CPU being involved (much). The "overhead" of USB would not account for such a huge performance loss compared to Firewire though.

X-Bitlabs USB2.0 test shows similar performance to your own with a "prebuilt" external drive on an In-System chipset.

My own enclosure uses an Oxford chipset for Firewire connection, connected to a Firewire card with a VIA chipset (both enclosure and card CompUSA branded, made by FMI). Sandra shows my scores not much better than your USB2 scores, and a lot worse than your Firewire scores (using a Maxtor D541X 40GB hard drive).

 

[JellyBaby]

There was some talk a couple of months ago about how the first USB 2.0 chips have subpar design work. Future USB chip designs might improve peformance but the nature of USB means it will never come as close to the theoretical bandwidth limit as Firewire.

 

[corkyg]

I have both USB 2 and Firewire 1394 in one of my systems, and Firewire in two more - no USB 2. USB 2 has a lot of problems as indicated above - all excellent and informative comments. On the non-tech side, I find Firewire to be much more user friendly with respect to devices and PnP. Now, 1394b or "Firewire II" is imminent and will probably hit the street with devices and cards this year. That will kick the data flow to a designed 800 vice 400, leaving the 480 USB in the dust again.

USB 2 reminds me of Zip - overly hyped by reviewers and tech media - like they were being egged on by Intel to do so. They hate Firewire because of "NIH." But, knowledgable users who think for themselves will probably try USB 2 but comeback to Firewire. It is a branch of SCSI, and is not new - mature technology - based on the separation of the circuitry that made SCSI 3.

It always makes me smile when I see how much good stuff comes to the PC from Apple.

1394b

 

[Eug]

Er... USB 2.0 and Firewire speed specs are for native devices, and only are theoretical speeds.

What you are using are IDE devices and require the use of a bridge chip. Right now reviews have the fastest Firewire chipsets (Oxford 911) hitting about 35-40 MB/s, which is about 70-80% of maximum theoretical speed, but the current NEC USB 2 chipset is probably only capable of around 15-20 MB/s in the best of circumstances, which is about 30% of maximum theoretical speed.

Newer production USB 2 bridge chips will have increased speed, but I'd be surprised if they ever surpass the best Firewire I speeds.

Thus right now for external devices, both from a speed standpoint and from a reliability standpoint, Firewire is the preferred technology IMO, partially because USB 2 is still such a young technology. Actually overall, even if USB 2 were more mature, in my mind Firewire is still the superior technology for a number of reasons, including greater supplied power, no CPU requirement, etc., and has a couple of added bonuses such as local peer-to-peer networking, etc.

By the way, Oxford Semiconductor has recently completed their Oxford 922 chip, which includes Firewire II support (with a max theoretical speed of 100 MB/s), and which includes USB 2 on board.

Oh and the X-bit test isn't really a good test, because it uses a laptop drive. It's gonna be slow regardless of the chipset. The interesting thing is that they confirm you can't run the laptop drive off of USB 2 power. Firewire can provide up to 15 Watts, which is sufficient to run off of Firewire power. This is moot with PC laptops though, since for some stupid reason PC laptops don't provide any power, whereas Mac laptops will provide the full 15 Watts.

 

[Lord Evermore]

Well, what I was impressed with the x-bit test was simply that the drive performed as well on USB2.0 as it did on a direct ATA66 connection, for the most part, and even with it being a laptop drive, it still did as well as or better than my own setup on Firewire.

The external enclosure I use has an Oxford 911 chipset in it. The IDE cable is also only a 40-pin, so I know it's running at best in ATA33 mode to the converter chip, and more likely in PIO mode (and I wish there was a way to check). I wonder how much performance difference a 7200RPM drive would make. The current performance is plenty for what I use it for, just as a backup drive and extra storage space, but it'd be interesting to compare. Too bad the enclosure is a pain to swap drives.

 

[LanEVENT]

Can someone that knows the differences between these two standards share why USB 2 most likely wont outpace FW even in maturity, like what overhead is there compared to FW.

On a side topic, I have heard 2 stories about FireWire - one that Apple "invented" it - another that Texas Instruments actually made it. I always thought that FireWire was just 1394. Was FW simply the first implementation of the standard, and who really did the work behind it? I know that all the FW chips I saw in the early days were TI.

 

[Eug]

Originally posted by: Lord Evermore
Well, what I was impressed with the x-bit test was simply that the drive performed as well on USB2.0 as it did on a direct ATA66 connection, for the most part, and even with it being a laptop drive, it still did as well as or better than my own setup on Firewire.

The external enclosure I use has an Oxford 911 chipset in it. The IDE cable is also only a 40-pin, so I know it's running at best in ATA33 mode to the converter chip, and more likely in PIO mode (and I wish there was a way to check). I wonder how much performance difference a 7200RPM drive would make. The current performance is plenty for what I use it for, just as a backup drive and extra storage space, but it'd be interesting to compare. Too bad the enclosure is a pain to swap drives.

Well, it's true that the USB 2 speed was almost as high as the ATA66 speed, but that's because the drive itself maxes out at about 18 MB/s, with an average of about 14 MB/s. Going to ATA66 doesn't mean it's going to run any faster. In this case the telling piece of info is the burst transfer rate. 58 MB/s for ATA66 and 20 MB/s for USB 2.

The key here is that if you hooked up a fast 7200 rpm IDE drive to this USB 2 chipset, the max read speed would still be around the same 20 MB/s speed. However, if you used ATA66, you'd be well over 40 MB/s for max read transfer rate. Firewire would be in the 40 range, slower than ATA66, but almost twice the speed of USB 2. But like I said, faster USB 2 chipset speeds are coming, including from NEC which is stating a 20% boost, and some other company I can't remember which is claiming an even higher boost. Neither can compete with Oxford 911 yet though.

Why your setup is so slow I dunno. Is your drive slower to begin with? I'm not familiar with that drive. Is it an older drive? By your comments and by the name it seems like the drive is a 5400 rpm anyway. And who knows, Firewire isn't perfect either so it's always a possibility that there might be some hardware incompatibility, although it seems to be working fine.

As for PIO vs. DMA, I was under the impression that Firewire was DMA by default but I don't understand the technology well enough.

Can someone that knows the differences between these two standards share why USB 2 most likely wont outpace FW even in maturity, like what overhead is there compared to FW.

Dunno actually, but I've read posts by several engineer types who believe the same thing. Maybe somebody else can expand on this.

On a side topic, I have heard 2 stories about FireWire - one that Apple "invented" it - another that Texas Instruments actually made it. I always thought that FireWire was just 1394. Was FW simply the first implementation of the standard, and who really did the work behind it? I know that all the FW chips I saw in the early days were TI.

According to everything I've read, Apple (working with others) invented the technology, and indeed, Apple does hold the licencing rights to it. If you want to use Firewire in your device, you must pay Apple $0.25. The word Firewire is just Apple's own term for 1394a. There used to be licencing costs for that too, but Apple has recently stopped charging for use of that term.

 

[kgraeme]

IIRC, Apple invented Firewire worked closely with TI to create the controller chip architecture. In that respect, TI probably played as much a role in the practical implementation of what Firewire became as Apple did in drafting the spec on paper.

 

[OneOfTheseDays]

yeah USB causes nothin but problems. I mean even usb 1.1 hasn't been completely ironed out yet. I actually have switched back to using serial ports for my keyboard and mouse because i find things a lot smoother on the serial ports.

 

[Lord Evermore]

You have a keyboard that connects via the serial port?! :-D

PS/2 keyboard and mice work perfectly well for me, I've yet to find any functions that a USB keyboard or mouse can provide beyond them.

 

[Eug]

Originally posted by: Lord Evermore
You have a keyboard that connects via the serial port?! :-D

PS/2 keyboard and mice work perfectly well for me, I've yet to find any functions that a USB keyboard or mouse can provide beyond them.

In defense of USB 1.1. I have a weird problem, and only with Windows XP (not 2000 or 98).

When I log into my account, and then switch users, and then log out of the second user, and log back into mine, I sometimes "lose" the serial mouse. VERY irritating. I've since bought a cheap USB optical mouse and have both plugged in. (My serial mouse I use for gaming so that's why I keep it.)

So in this case USB has saved me.

 

[Lord Evermore]

Still...you use a serial keyboard?!

I found a serial mouse in my closet a few days ago. Forgot I'd stuck it there, hadn't used it in like 2 years but kept it just in case (it had actually started having tracking problems that cleaning didn't help).

Oof, I found why my external drive is so slow I think. It uses the Oxford 900 chip, not the 911. The 900 onlys supports up to ATA33 (it does support PIO modes but of course will use DMA modes if the drive supports it). The burst rates only go up to 50MBps, and sustained is only supported at 25MBps. An ATA66 cable of course didn't change the performance at all (I tried).

So, I'd suggest anybody buying an enclosure for Firewire, be sure what chipset it uses.

 

[LethalWolfe]

On a side topic, I have heard 2 stories about FireWire - one that Apple "invented" it - another that Texas Instruments actually made it. I always thought that FireWire was just 1394. Was FW simply the first implementation of the standard, and who really did the work behind it? I know that all the FW chips I saw in the early days were TI.

Just to add to what others have said. Firewire is Apples name for IEEE 1394 (which Sony calls i.Link). But recently Firewire offically became the name. No more Firewire...er... i.Link...er... IEEE 1394 name juggling.


Lethal

 

[LarryJoe]

Your first mistake here is using Sandra to bench your drives. It is well known that Sandra is completely unreliable for benching drives.

Try something else like hdtach or winbench.

EDIT - spelling

 

[McCarthy]

So if Firewire is 'official' now, what's 1394b? Firewire II? Firewire+? FIREwire? I kinda like the b

--Mc

 

[Lord Evermore]

Apple has stopped charging the fee to use the Firewire name, so the 1394 Trade Association has adopted the name officially. This is supposed to help push Firewire more, so that people aren't wondering whether their i.Link device works with their Firewire device, things like that. 1394b is Firewire II, though I'm not sure that's definitely official.

 

[ScrapSilicon]

USB 2.0 LTR 32123S(originally a LTR 16102B.. ) data backup CD(on el cheapo SpinX from OM..) with a size of 699MB time from start to finish using Nero...4:00 minutes flat...same time as out of the enclosure..CPU usage (with 15 IE windows open..) hit 20% max. Not a harddrive but .. serial ATA isa comin' ....... edit...too many yellow thingies..

 

[Eug]

Originally posted by: ScrapSilicon
USB 2.0 LTR 32123S(originally a LTR 16102B.. ) data backup CD(on el cheapo SpinX from OM..) with a size of 699MB time from start to finish using Nero...4:00 minutes flat...same time as out of the enclosure..CPU usage (with 15 IE windows open..) hit 20% max. Not a harddrive but .. serial ATA isa comin' ....... edit...too many yellow thingies..

A) USB 2 is fine for CD-RW and DVD-R drives. There is no bottleneck here because compared to hard drives, both technologies are dead slow. The bottleneck comes when you're talking about hard drives.

2) Serial ATA seems great for internal devices, but not well suited for external devices. You need an extra port for every device, and it provides no power among other things..