Lucmove: I just returned my drive to seagate for a replacement one. When I get my replacement, I will do some speed testing on windows/linux and post my findings if anything magical happens, but I have a feeling that my drive on linux will be slower than on windows, just like many of the people on ubuntuforums are reporting. I'm crossing my fingers though.

I wonder if my new/refurbished drive will be up to normal speed with firewire. Before, both firewire and usb were equally slow.

acgarib,
I just googled "Seagate Freeagent Pro" and saw that it has eSATA. Use it. It's a lot better than USB or Firewire.

Yeah. All I know is that a fresh install of Hardy Heron doesn't have this problem. It reads and writes fast.. Since we all know that a GUI isn't a reliable source of information, and that the gnome developers just threw those timers/speed readouts in to fool all us non-developers, I will say that writing files to a SimpleTech 320GB external HD takes around 20s per Gig before updating the default kernel to the current version, where 6min per gig is the new average. I verified this with the wholly unreliable and fictitious output in my GUI, a stopwatch and a sundial - just in case the people at Timex lie about the watchs' accuracy.

I know this post here is getting absolutely nowhere. Posts at Ubuntu Forums also come up dry because not everybody has the problem, or have the same attitude that, "well, it's working fast enough, so I guess it must just be how it is."

Where do kernel developers hang out? How does someone as ignorant of kernel workings and design as me get with one of these folks and talk shop?

lucmove: sorry, my laptop does not have an esata port. Besides, everybody that has used the freeagent pro via esata has reported problems (windows and linux). Apparently, the enclosure is slow and unstable when connected to esata, but some people have said that a firmware update has helped. Others have removed the drive and put it in a different enclosure and reported normal speeds.

I suppose I could buy an esata ExpressCard, but I don't think the extra speed is worth it considering I have to buy an esata expresscard and either pray that the firmware update works or buy a new enclosure and void my warranty.

I'm just hoping this gets fixed soon so I don't have to use windows for major backups.

I know this will be a pain to do but have you tried installing another distro to see if the problem still exists after updating to a new kernel?

I personally suggest fedora for relative easy install and relative big difference.

If the problem still exists the problem is probably in the vanilla kernel otherwise it's in one of either redhat or ubuntu kernel patches.

Trying debian would make this easier as well 'cous ubuntu is based on debian. So if there's a difference it's easier to identify.

I mainly run Gentoo and configure everything myself but have you tried running 'make menuconfig' on the older version install and checking the device setting for your usb ports and comparing them to the updated kernel settings.

As for speed checking i just use a LARGE (DVD movie usually) file to compare speeds using a stopwatch comparing transfers between the two systems with the same file. If it takes 3 mins with the first kernel but significantly more with the new one might need to play with the kernel settings or at least make sure the new kernel is running the new usb version support and not usb 1.0 or something. Not sure but it seems that is what you are doing just make sure its the same file you are moving so the times should be close.

As far as the difference in speed with windows vs linux I am not sure. Never checked ( might now though ).

Same thing here: performance on external USB disks has significantly degraded from Ubuntu 7.04 to 7.10 and is becoming unbearable in Ubuntu 8.04. This does not happen with all hardware though. One 500Gb disk would run fine around 20-30Mb/s and the other would peak at 3Mb/s and then drop performance after a minute or so. All devices run at full-speed on Windows.

This is not related to filesystems (same behaviour with vfat, ntfs, xfs), device driver mismatch (all using Spd=480) or mount options (all async and noatime). I also noticed recurrent filesystem errors occurring on drives that have slow transfer speeds. This would need finer experimentation but it looks like a regression in the USB code in the kernel.

Just to set things aside: we are not talking about a 5-10% decrease in performance but a factor 10. We do not need fine-grained measurement tools at that point but a systematic approach to find out what is going on. Anybody knows how to get in touch with the USB guys for the kernel? I would love to have low-level measurement tools and logs to try to capture what is happening there.

+1 for that. I asked sorta the same thing a number of weeks ago and never got an answer out of anyone.

Uncertain,
I hope you found the answers you were looking for. I find it completely ridiculous of the replies from Electro. Apparently he has some personal interest in Firewire technology and is trying to push it. While Firewire IS rated faster, I consistently get 25-30 Megabytes/Sec transfer to my USB drive (SATA HardDisk in a USB enclosure).

As of this writing, I am transferring 110.7GB of data from an internal disk, formatted with EXT3 (SATA internal), encrypted with TrueCrypt (Double encryption as it is on a hidden drive, being copied to an NTFS formatted drive (USB). Take into account, that the encryption creates overhead and I am still achieving 26MB/Sec+ transfer rate.

I find it appalling that your question kept being avoided and being told that USB and Linux has issues. My guess is that the other guy is a MAC user, where firewire is more popular. We use firewire here at work on a MAC Server and to be honest, it doesn't appear to be all that much faster. The rates of speed that he kept talking about ARE true (WITH USB 1.1!!!) USB 1.1 is max rated at 12Mbits, which converts to 1.464844 MB/Sec. Apparently he needs to educate himself on USB 2.0 speeds vs. V1.1 speeds.

Linux is actually much faster copying files than Windows ever has been. I've booted Linux FROM a USB drive and never really seen much lag time, or slow performance. At least not like I did when I tried to install Windows on a USB drive. That was painful.

It is my guess that you either gave up or found the answer to your issue. If not, try to update your system to all the latest updates and see if that fixes things. I am using 8.10 Ubuntu with the latest updates and now sitting at 26.3 MB/Sec. For those who think the GUI speeds are NOT accurate.... Hmm, I can count 1-1000, 2-1000, etc. and the actual time to transfer X-amount of data IS accurate. As for USB overhead. Sure, there is. USB 2.0 is rated at 480 megabits. That converts to 0.05859375 GB, or 58.6 MB/sec. The BEST (actual) speed I have ever achieved from a USB drive was about 39MB/Sec sustained rate. That's copying a 4.7GB .ISO file from the USB drive to an internal drive.

-Cybie

Actually, no - I haven't. Unless you consider re-installing the OS and locking kernel versions so they don't update past the default install a solution.

There seems to be a consensus over in this thread at ubuntuforums that it's a kernel issue. Maybe it's electro in disguise , but there's one fellow over there who is insisting I have a hardware issue because I noticed the slowdown applies to copying/moving files between partitions on the same internal HDD.

For me, with my particular hardware, I don't have any problems with a default Hardy Heron install using kernel 2.6.24-16 and it's associated modules/headers/whatever. As soon as one update is performed and a newer version than that comes into play, it's slowville all the way. Some people say that Intrepid holds the answers, some people say that Jaunty has a fix... For me, anyways, both those versions are just as f'd up (if not, worse). As a matter of fact, until the most recent kernel updates, my read cycle (copying data from a USB device and writing it to my HDD) was unaffected - I had full speeds approcahing and sometimes hitting 30MB/s. After the last couple updates, I now have the same slow speeds when getting data off of a USB device of any kind.

In any event, I might as well publicize this here, too. I've made the offer in that thread and now here that if anybody more knowledgeable than I wants to see this bug in action, I can do a re-install and then update my system so that it goes off. I've been meaning to do a re-install as of late, and if someone wants to watch this monstrosity unfold, it would be a good incentive for me to do it. I'll provide VNC or SSH access if needed, but would gladly post any command outputs necessary to narrow the problem down and see what causes this.

I don't know if there's any developers or USB or kernel gurus that read these forums, but the offer's there.

I have never gotten 30 MB per seconds for USB 2.0 and yes I have USB 2.0 hardware. I know the difference between USB 2.0 and USB 1.1 speeds. I have not gotten past 10 megabytes per second and this includes the file system. Recently the reliability of USB with the latest kernel has gone down hill. What is ridiculous is people are not stating what hardware they really have. Another funny thing is people that have the knowledge are complaining about the poor performance of USB in Linux but do not have the time to fix it. If Intrepid and/or Jaunty provides better USB performance for USB, prove it.

IEEE-1394 uses DMA for data transfers while USB does not. Microsoft implements DMA for USB even though that is out of spec. Linux does not use it. Linux methods is like PIO, so that is why it is very slow. Also wrapping around with SCSI commands makes things even worst.

I use IDE removable drive bays to do backups and to transfer over because USB overall is pathetic for data storage devices. My future computer build will be using SATA and I will be using that connection instead of USB. SATA provides the same features with out the stability and reliability problems of USB. Of course SATA provides more bandwidth than USB. At that point I will not be using USB.

The cut and dry solution for USB in Linux is completely remove the USB infrastructure altogether. Then write a whole new USB infrastructure at a different perspective in assembler. Sure C could be used, but it will be slow. Linus will not like this because the assembler code have to be rewritten for each processor type (PowerPC, ARM, 80x86, etc). I would like to do this solution, but I do not have the expertise nor I have the money to spend on a rig to test and develop the code.

IEEE-1394 or Firewire is not just for Mac and I do not own a Mac. It is sad that a lot of people think that IEEE-1394 is meant to be used for a Mac but this is wrong. IEEE-1394 is better than USB, but I have seen a lot of companies cutting corners using low grade chips that makes IEEE-1394 perform worst and/or be unreliable for the IEEE-1394 connected device.

Complaining about the problem does not fix it. Also using a different connection does not fix the problem.

I am using Gentoo. IMHO, Ubuntu are not that experience as Gentoo, Slackware, and Arch users and it shows. I guess Ubuntu users does not know that USB can easily be set in debug mode, but the kernel may have to be recompiled to include the ability to do debugging. I mainly compile my kernel with debugging for USB because USB for me is always unreliable and I want to see what is going on.

It'd be interesting to see the output of a diff between the two kernel configs. There might be an option that has been changed which is causing this problem.

Tip: You don't need to re-install the whole OS just to revert back to an earlier kernel. Ubuntu doesn't delete old kernels when an upgrade is applied, so reverting to an earlier kernel version can easily be accomplished by editing your GRUB configuration. This will also enable you to determine that the problem is actually caused by the kernel.

Also, upgrading the kernel on a "perfectly healthy" setup is not always a good thing, unless there is a security fix or some hardware support that is needed in the later version.

0 members found this post helpful.

I've tried that - it doesn't work. Once the "fatal" update gets performed, it ruins all kernel versions you have installed. (At least for me - others might have a different experience.)

I have only my default-install 2.6.24-16 version and the latest Hardy version (2.6.24-23) and both kernels have the slowdown. Otherwise I'd just always run in .24-16 and wouldn't worry about it ever again.

@Electro:
Feel superior much?

If you've never gotten much faster than 10M/s out of USB2.0, then you have the same bug as everybody else and don't even know it.

Odd. Very odd. I might be wrong, but this would indicate that perhaps something else is at fault? What else gets updated in a "fatal" update?

Could you upload the "broken" kernel config (and a known good one to compare it against) somewhere and post the link here so that I can take a quick look when I get home from work?
uncertain: I feel your pain, but at this point I'd be ignoring this guy. He has contributed nothing useful to the thread so far and probably won't. Don't waste your time or energy.

Umm, I think I did by stating that I was getting 26.3MB/Sec while writing that last reply. And yes, that was the nautilis GUI progress bar/status time shown, but the MATH showed it to be accurate within +-5%.

As for hardware:

GigaByte Motherboard GA-P35C-DS3R
4 GIGS Kingston 800 DDR2 RAM
Core 2 Duo Intel CPU 1.6GHz (Speed may be slightly higher)
On Board USB to Externeral USB Hub (Hub = Cheap Generic 7-Port, powered)
EverTech (generic brand) Ext USB 3.5" SATA HD Enclosure
Cheap generic USB 3-foot cable
(Other hardware irrelevant to USB topic)
Ubuntu 8.10 All the latest updates (No new updates available at this time)

Portability, compatibility, USB is the way to go. Even if it's not the fastest. Yes, I like E-sata, but what happens when I take my external HD to a friends house who doesn't have E-Sata? It like saying Blu-Ray is the way to go. Yeah, 50GB is awesome, but what good does it do if the majority of people don't have the ability to read the medium???

I never said it was just for Mac. I said it is more popular for Mac. It was developed by Apple to replace the parallel SCSI bus. Firewire also is being held back due to royalty requirements by Apple. That's why it's mosty found in MACs.


The problem with your replies aren't that you don't have any less right to speak your mind, it's that you avoided the original question of this thread over and over. The question was, USB was going "THIS FAST" and now it's going "THIS SLOW" after updates. Hardware is NOT the issue. I achieve 25MB/Sec on Average copy to and from my USB drive. If you don't believe it because yours doesn't work, then great. Mine does and I have had cheap motherboards in the past that have produced poor USB performance. But then again, the entire system was poor. Your computer will only perform as good as the hardware is. In my opinion, based on experience with numerous manufactures, GigaByte provides one of the best, affordable motherboards out there. In fact, the board I have will even run OSX. Yes, I have done so already with no special hardware. Search the net for the board model and OSX and you will find it on the list to run OSX natively.

Oh, and just an FYI. Copying from a native EXT2 Internal SATA to the same external drive (formatted as a Native EXT2), provided read/write speeds at and above 35MB/Sec. That's with nothing else being done on the computer and is a fact.

-Cybie