[OT] Connect external hard drive to internal SATA port?

[Paul]

Chipset ID Renesas/NEC µPD720202

At least it has good odds of having a WinXP driver. NEC made the first USB3 chips (so the range of OSes supported is better than some others). I use NEC on this machine (running WinXP) for that reason. But mine is sitting on PCI Express bus (not a bridged NEC card like the Amazon link above), and I get a decent but not dazzling speed (200MB/sec).

But the typical usage characteristics of PCI-bussed boxes, means it's hardly worthwhile. You'd be getting

50MB/sec operation to the external device, during typical backup conditions. Half the bandwidth used by the internal drive (on IDE), half the system bus bandwidth available on the new USB3.

CPU | <=== choke point on this one NB | PCI bus <=== skinny choice, 110MB/sec with high burst setting | +--- SB <=== Southbridge is a PCI client | +--- PCI to USB3 card

The PCI bus is a bottleneck for storage operations.

The Athlon wasn't necessarily a bad processor, but the chipset architecture is part of that "PCI everywhere" thing that dooms the boxes to be starved forever.

Even the system bus from the CPU to NB runs slow. If you overclock an Athlon (I tried it), it offers no benefit whatsoever. It just means the CPU burns more cycles waiting for memory accesses to come back. I was a little disappointed with mine, and it only stayed at high clock for ten minutes, before I set it back to stock speed. A waste of time.

If I'd spent more time overclocked like that, I might have noticed the timekeeping on the system was screwing up. The NForce2 chipset had a bug where non-canonical bus clock choices tend to "wander" on real time clock. No amount of NTP syncing of clocks does a bit of good. Someone turned his up to syncing once every minute and it didn't help. And I overclocked for a short enough time, I never even noticed that.

Paul

[Pamela]

[Roger Hayter]

That's eSATA, but it is not very common.

[Pamela]

Compression is not an answer as it soaks too much cpu which often gets maxed out.

The tested R/W speeds don't entirely make sense. The software gave one popup glitch which may indicate it's not working well. Subjectively, 50GB backups feel slow and read/write of data files is less than crisp.

CrystalDiskMark ver.7 with 256 MB test data (all in MS/sec):

ext Seagate "Backup Plus Slim" seq: 19.5R/19.5W, random: 0.4R/1.3W int Hitachi "Travelstar" 7200rpm seq: 114R/97W, random: 0.5R/0.4W int Samsung SSD EVO 850 seq: 118R/100W random: 15R/13W

[Paul]

So it's a S754 or S939 then, and not an original Athlon S462.

The arch on those is a bit better.

CPU --- RAM | <=== Hypertransport bus VID----NB | <=== Hypertransport bus (lower BW than bus above) | SB --- disks | PCI bus Now, the disk bandwidth does not compete with USB3 card | +--- PCI to USB3 card Backups at 100MB/sec approx.

If the caps on the motherboard have held up this long, they're probably not weak ones. I've not had any motherboard failures here due to caps. And that's not from careful selection, just luck.

Paul

[Paul]

The only USB2 this slow, was on SB400 Southbridge from ATI. The SB400 was noted for doing its USB2 at 20MB/sec for some reason.

ext Seagate "Backup Plus Slim" seq: 19.5R/19.5W

Most other USB2 should be able to do 30R/30W at the very least, with the occasional situation running at 35MB/sec (out of a wire rate of 60MB/sec max).

Paul

[Pamela]

[Pamela]

This Asus board has a GeForce 6150 LE chipset. It's still slow.

The internal Hitachi 7200 random read-write performance seems surprisingly slow and is no better than the ext USB drive. That's why I wonder if the benchmark software is glitching.

Maybe the external USB hard drive's electronics is maxed out because the read test gave 19.50 and the write test gave precisely the same 19.50 value. I would have expected a tiny chance variation.

[Fredxx]

Do you do a complete backup, or only the files that have been touched.

How many files, what size in total and how long?

The sequential speed is 1/3 of USB max speed, so any improvement is not going to be significant.

One of the reasons I moved to Veracrypt is the files are stored in an archive; where the FAT is part of the archive and the read/write speed for small files is very much improved.

I use FreeFileSync that detects files that have been touched and let things run overnight. It can also have file versioning for files that have been overwritten.

[Paul]

I have an Optiplex here and the thing uses plastic trays with a kind of spring action to hold the disk drive in the tray.

It works, but I would not want to be rotating drives on a daily basis using those. Being plastic, you'd need to stock twice as many of those in case one snaps.

If you bought enough trays so every drive had one of those trays, and were sliding them in and out, then that would be great. But multiplexing one tray with a dozen drives would be a kind of hell.

The card retention on the card slots is delusional. Does not inspire confidence, like the screw down stuff in the other machines here. I'm always fearful a card will fall out or something.

And that's all the good things I can say about toolless design.

The Optiplex itself works fine. Quiet. And part of the reason it can be quiet, is the overall machine power consumption is 65W at idle. (Both my home built machines are higher than that.)

I think a quad core refurb would be a better choice, in terms of a bare minimum for Windows 10.

Things I've tested:

single core laptop - "operational", but a bit on the torturous side

dual core Optiplex - better than the single core machine. Will still be times I would regret using Windows 10. The use of DDR3 in the design (versus DDR2 on the machine I'm typing on) seems to help.

six core MB - Now, I'm needing more straight-line clock speed. Maybe 5GHz would feel better than 3GHz etc. Does everything I could want. Windows 10 always makes me wonder though, where all the CPU cycles are going, and if you use ProcMon, there's an unbelievable amount of background activity.

Just the time it takes to install stuff (whether using SSD or not), where are all those cycles going ? A lot of this traces back to their WinSXS maintenance concept, package management (the OS is a thousand or more small packages, glued together). Windows 10 is "mostly tax" in a sense :-) That's why you're going to consider my description of performance requirements to be puzzling, because the above isn't like other OSes where "the machine is yours".

It's the same as if you used Snaps on Ubuntu. Lethargic startup, because of delusional concept. They say it's progress and everything. Russian dolls computing is the future.

Paul

[Paul]

Try HDTune (read benchmark). This does a sustained read bench.

These are some results, on various USB controllers. USB2 gets 33MB/sec in the example. None of the results are world records or all that impressive. This is to show real world results using random items. I also have a USB stick that runs at 1MB/sec... but you wouldn't be impressed.

The Asmedia with the USB3.1Rev2 was not selected for that standard. I bought that expansion card, because it has two PCI Express lanes (x2), and can make USB3.0 devices run a tiny bit faster. I don't have any USB3.1 Rev2 peripherals (you get those in the form of USB to NVMe adapter boxes). A USB to NVMe adapter can do 1GB/sec, if connected to a native USB3.1Rev2 port. Mine will never run that fast. Even if I could afford something to plug in like that.

If you put a USB3 and a USB3.1 storage device on a USB3.1Rev2 controller, they don't play fairly with one another, and one of them will end up with inferior transfer speeds. That's for the rich people to figure out.

Paul

[Pamela]

[Paul]

The Optiplex tower is not a "tank".

That toolless crap is why. Anything toolless, some day it's going to snap on you. Which is fine if you can afford spares, stocked in advance of the event.

The parts of the machine that stay put on the Optiplex are OK. I managed to take a DVD tray cover off, and now I can't figure out how to put it back :-) How embarrassing.

Paul

[Pamela]

I do it differently by backing up the entire system partition and keep its size down by not storing user data there. I don't create a cloned image file but a fresh partition which a boot manager can boot from if I choose.

The slow external drives are not for backup but data storage. I figured on better performance by attaching them direct to an internal SATA port.

[Paul]

The problem is, those drives, when you pull them out of their enclosure, only have a USB3 connector on the disk controller board. And that was done specifically to stop people from using the drives in SATA applications.

Paul

[J. P. Gilliver (John)]

On Mon, 8 Jun 2020 at 21:43:43, Paul [e-mail] wrote: []

Why, though? External drives in a housing, with a cable, were - at least initially - more expensive than the same drive as a bare drive, so why would anyone _want_ to?

[Paul]

It's a lot of effort, to stop a few Ebay sellers. Making them native USB3 just so they can't be sold on as tiny SATA drives.

Paul