Anyone know of any recent reviews of USB memory sticks for their speed? The quoted speeds in retail seller sites is worthless because they are just citing how fast it can dump a sector of data across a USB 3.0 port. And the reviews I have seen fail to differentiate sequential vs. random write/read orders.

It would help to have some good standardized tests for comparison purposes. But one thing to be avoided (or at least set aside as a separate test) is reading/writing files in filesystems. IMHO, those tests are hard to reproduce. A random sector test should synthesize the worst case examples. Maybe a mix of random and sequential could be reasonably close to filesystems.

I often do backups to these devices with a tar or cpio command, maybe with compression, or maybe not (if I know the data is mostly already compressed as compression does slow things down). For example "tar bcf 1 - | vbuf -o /dev/sdc1". In other cases other backup means can be used, such as rsync when doing incrementals, and that would be more random, depending on what files need to be updated.

BTW, I use ext2 on these more often that fat.

In other cases I build bootable images and copy those to the device sequentially. That's effectively like the tar example above, with "vbuf -i bootable.img -o /dev/sdc".

The vbuf program is available in the Ubuntu repository. It might also be in some others.

So the final question is, where is some good technically valid info in available device speeds for USB memory sticks? I'm looking to buy a few in the 8GB range for making some bootables, and a couple 64GB for portable backups (maybe also bootable).

The "transfer speed up to" figures marketing people use is worthless.

I bought a well reviewed (by windows users) usb 3.0 flash. It worked great on windows. Super fast and just what the specs said. So, I thought I'd load on a linux distro to it native ext4 format. I tried and tried and never could get it to work worth a darn. So, unless someone posts an exact linux install or some exact test like you need then there is no way you can guess.

I felt kind of bad sending it back too. I tried different formats and you name it. It just didn't seem to like to be booted to a linux OS.

Interesting. I've never run into that problem. Too bad you sent it back, as I would have liked to have tried some suggestions on making it bootable, or at least to benchmark the sequential speed of it.

What was the symptom? BIOS refused to try to boot it? Bootloader crash? Bootloader not find the kernel/initrd? Kernel not find the root filesystem? The rc scripts doing something wrong? Was Slackware among those tried? I have had some difficulty with Ubuntu when following their documents on making a bootable USB. But I have had good success at morphing the Ubuntu desktop ISO images into a hybrid image that still works as an ISO (e.g. you can burn the result image to CD/DVD and still boot that) and as a drive image for any flash drive or hard drive. It works because the Casper program used to bring it up will look even at hard drive devices for its root image. Also, I have a project I'm working on to make customized Slackware based boot images for flash drives. I even have dual 32/64 versions working (you choose which to boot at bootloader menu). It uses Syslinux plus a program of my own to find which device node the drive shows up at (since that is inconsistent between different machines).

Booted OK but a bit slow. When it ran it would run fine for a bit then freeze. Seemed to me it was some controller getting in some loop on the usb. Some kind of race condition.

What sort of USB memory sticks are you looking for reviews on? USB 2.0, USB 3.0, USB/eSATA?

USB 2.0 tops out at about 40MB/sec, and 30MB/sec is the typical USB 2.0 speed. USB 3.0 or eSATA flash drives will be a lot quicker, but only if you have the right ports....run them on USB 2.0 ports and they drop to the normal USB 2.0 speeds.

Was there swap space?

I've gotten 45MB/sec on USB with hard drives on USB 2.0. Finding a USB memory stick that can do 40MB/sec and sustain it over the entire drive capacity when doing sequential write or read would be great. A good review tests it both ways (random and sequential).

USB 3.0 is of no interest until they get the flash bits fast enough to make USB 3.0 meaningful.

I have no interest in eSATA at the moment as I have only one or two machines that can do that. When the ports become ubiquitous then I will have interest. Same for Firewire (which I do have a few ports for but never seen memory sticks for).

I do have CF that can do 60MB/sec, and seen marking for ones that do 90MB/sec and more. But CF has the advantage of 8 data paths. Then there's C-Fast (eSATA in a CF form factor) which I have seen NO media for at all.

Tried it with and without swap. Pretty sure it was an issue with the usb flash controller.

Possible, fairly uncommon. USB 2.0 can sometimes get to 40MB/sec, and very very rarely can break that. But USB 2.0 shouldnt be counted on to get more than 30MB/sec even if the device is capable of far more.

*edit-

http://reviews.cnet.com/external-har...4183942-2.html

They have. There are various flash drives that are faster than USB 2.0 bandwidth, even if you believe that USB 2.0 is capable of 60MB/sec. Eg, Kingston DataTraveler Ultimate 3.0-

http://www.legitreviews.com/article/1668/1/

http://www.tomshardware.com/reviews/...rive,2900.html

1 members found this post helpful.

So how about one that does the full 30MB/sec read and write on USB 2.0, but does NOT go way over that on USB 3.0 so as to keep the cost down for the USB 2.0 market (which still still be around for quite some time)?

Even brand new machines bought in the last year have no USB 3.0. I don't want to overpay just to get the maximum out of USB 2.0. And I'm not holding off on machine purchases to wait for 3.0.

Maybe it is the case that the very tiny USB 2.0 controllers they put in these can only do 30MB/sec. I only saw the 45MB/sec speed with a hard drive type device, not a flash memory stick. But apparently someone does know how to push USB 2.0 to the limit. But with 3.0 out, maybe they (management, not engineering) decided to back off (it might be an expensive chip) since 3.0 will probably be ubiquitous in a few years (probably cheaper to do 100MB/sec in 3.0 than do to 40MB/sec in 2.0).

AFAIK the USB 2.0 drives that were pretty quick (eg Corsair flash voyager GTR) have pretty much all been deactivated/removed from the manufacturers inventories.

The manufacturers know that they can make a slightly faster USB 3.0 version and charge a premium for it.

Depends on what hardware you buy. The 'corporate' entry level and even mainstream level systems you can still find with USB 2.0 only. Thats just corporate pricing and tech uptake for you, USB 3.0 is a feature that is still seen as something that only enthusiasts want. And/or something that they can charge more for.

For desktop machines a PCI/PCIe USB 3.0 card is cheap, about $20 (PCIe) or $30 (PCI). Doesnt help with laptops though.

Its just a pity that the eSATA/USB2.0 drives appear to be deactivated/removed from the manufacturers inventories as well. They were generally quick on USB 2.0, blasing fast on eSATA, and a lot of chipsets/motheroards let you change an internal SATA port into eSATA (with some hardware).

http://www.pureoverclock.com/review.php?id=802&page=1

If you can find a eSATA/USB 2.0 drive, it might still be an good option.

Even 45MB/sec isnt really pushing USB 2.0 to the limit.....if you believe that USB 2.0 is capable of the full 60MB/sec. I dont (and neither does the USB-IF )-

http://en.wikipedia.org/wiki/Universal_Serial_Bus

Its also half-duplex, which makes a difference. Plus, its limited by other devices using USB.

I still believe that anythign over 30MB/sec from USB 2.0 is doing very well, no matter what drive you are using.

The 60MB/sec is only the bit rate transfer speed. USB's poorly designed protocol (too many turnaround steps that have to wait) makes that impossible to reach. I don't know the details on how USB 3.0 improved this. So it could be removing some of the protocol bad points, but it has to be a higher data rate to get above 60MB/sec, anyway. A better controller might know tricks to minimize the protocol wait steps and make better utilization of whatever bandwidth there is. If they didn't improve the protocol, then even 3.0 will see these issues when it reaches half its data rate speeds sometime in the future (e.g. devices above 300MB/sec will start to see the headroom limits).

If you are just doing reading or just doing writing, it would not make that much difference (depending on the maximum block size or maximum number of chained blocks). If the protocol allowed back to back block transfers, you could get very close to 60MB/sec (as long as the device itself could deliver that fast).

If it were full-duplex, that would be done by either another pair of data wires that could, instead, be used in half-duplex usage to double the data rate. And if they change things to do full-duplex over a single pair (very doable), they could just as easily do multi-level encoding, too.

If they wanted to have more wires, they could also do multi-wire encoding.

75% is definitely "very well" for USB. I don't know what tricks were used to get there, but I know it has been done since I saw it. 75% is considered "it sucks" on ethernet (I can do 96% on ethernet). I've heard 80% is not hard to do on Firewire, but have not tested it. I don't know much about [e]SATA nor do I have something fast enough to see where its limits are. I do know that there remain issues with hot-plugging eSATA but I don't know if that is a standards issue or an implementation issue.

edit: I just saw that USB 3.0 went to "twice as many wires ... to support SuperSpeed". I wonder if it allows both pairs to go in the same direction for large fast transfers.

It comes down to:

1. The USB 2.0 needs are still widespread.
2. The speed in USB 2.0 is very dependent on the USB 2.0 implementation/design.
3. So we still need to have the faster devices listed as to their USB 2.0 speed.

A device very fast on Flash chips, and this fast in USB 3.0, might well skimp on USB 2.0 controller logic, and be among the slowest in USB. These are devices intended to be portable, so they need to have specs that address a portable reality.

That is the basis of your problem....these days pretty much all the 'fast USB stick reviews' are done on USB 3.0 controllers. Not made any easier by the general USB issues, the number of different controllers, and the way that testing is almost always done on windows.

IMO you'd probably do best to see what is avaible, find some sticks that are in your price range, then go looking for reviews/benchmarking.

The few times I've seen people try to do it 'backward' (find reviews, then find the stick) they either have problems finding those sticks, or they get something that has the same naming but can be a different product (you probably already know this, but supermega!32GB-12345678 can be quite different to supermega!32GB-87654321, and a lot of places leave the '87654321' bit out), or they end up paying more than they would have otherwise.

And of course it's just not the same as the good ole super!mega32GB-12345678 :-(