"Download to D:" bug persists :-((

[tomas]

I hoped that the buggy slowdown of download to D: drive with Opera might be corrected in the new OS&Opera, but it is still there. If downloaing file to C: from internet, I have got some 100-150 kB/sec. While downloading to D: the same file from the same server at the same connection, speed drops to some 10-30 kB/s. CF is slower than RAM, but still much much faster than these 10-30 kB/s, so it cannot be a media speed problem...

[wanman]

Mmm. Those speeds do seem rather rubbish. What is the size of it / manufacturer? Have you tried Disk Bench to enable you to compare times with others posted. Check out this web site
http://www.fortunecity.com/skyscraper/babbage/399/

It is true that Opera definitely prefers storing it's cache on the c: drive so I would try and use that if you have the room. When I have tried D in the past it has been noticably slower but I don't think as noticably as you are reporting.

Hope you manage to sort it.

Regards

Si

[tomas]

I use 1GB Microdrive. I performed several tests - write speed under epoc is 190 kB/s.

[tomas]

Over the local network, opera can give even more speed, over 300 kB/s. But only to c: drive...

While testing download with nFTP, it seems to have some software limit around 80 kB/s, but gives the same speed to c: as to the d: drive...

[diem]

just to clarify TOmas - does this slowdown occur with just the Microdrive or with non-volatile RAM CF cards as well?

[tomas]

I tried also 64MB CF with the same results :-(

[wanman]

Apologies Tomas,

Not able to help any futher. I have not come across this before, even when I did use a CF for my Cache folder. Sounds like a bit of a stange one. Do you get write problems to your external drives with any other applications other than Opera?

Si

[tomas]

No, the problem seems to appear only with opera. How can I redirect cache to d: ? I would try if it changes something...

Slowdown problem is noticeable only when connected to ethernet. Through slow modems there is no bid difference...

I would be glad if somebody could try a little test too:

1. When connected to ethernet, download some 10-20 MB file with Opera to drive c: (find a server which will give you at least 70 kB/s, so we can see the difference).

2. Immediatelly download the some file from the same server with the some connection again to d: drive

3. Post here both reported speeds from transfer window

Thanks...

[tomas]

Pretty please :-) Could somebody with ethernet connection try this out ???

Many thanks...

[netBookBabe]

Sorry, tomas, got the card but haven't got the connection yet!

:(

Julie

[wanman]

Sorry Tomas, got the connection but not the card....Doh

Si

[netBookBabe]

I'm thinkin' that Si and I should get together! ;)

ROFL!

Julie

[diem]

I'll do the test tonight.

[tomas]

I have just performed another interesting test: The current version of nFTP (inbuilt) gives numbers different from previous version of nFTP, but very similar to the Opera... So now it seems it isn't Opera's problem, but OS problem...

I downloaded 30 MB file from local FTP:

to the C: Drive with 118 kB/s

to the D: Drive (64 MB CF) with only 14 kB/s.

There has to be an OS bug !!!


And how was your testing, diem ?

[martyscholes]

Tomas,

I discovered this very thing the other day. I keep an old linux box around for ripping CDs for my CF MP3 player. Unfortunately, the CF support under linux is weak. I rip the tracks and then download them to the netBook and use it to put them on CF.

I downloaded a larger file (27MB) to C, then copied it to D. I got good speeds going to C, about 400KB/sec. I then decided to "cut out the middleman" and download the next large file directly to D. I got terrible speeds. IIRC, it was about 15KB/sec.

I tried it both with opera and nftp and got similar results with both. Seems odd to me.

Marty

[diem]

Using an 8.5MB file.

nFTP to C - had dropped to about 107KB/s by end of transfer.
nFTP to D - stayed around 15 KB/s throughout transfer (killed it 'cos I got bored!).

Opera to C - had dropped to about 292KB/s by the end of transfer.
Opera to D - don't know 'cos it wouldn't update the screen whilst downloading - the file size grew when viewing in System, but Opera itself was unresponsive, stuck displaying the download dialog. It was quite clearly infinitely slower than downloading to C however.


A couple of things to note:

Opera is faster at downloading than nFTP - I've looked into this before now (against RMRFtp), and nFTP is definitely just slower. I've emailed neuon to no avail - I guess their routines aren't as efficient as Opera's/RMR's. Tomas, you mentioned that the numbers have changed between your versions of nFTP - what are your version numbers and has it sped up or slowed down?

Does anyone have any proof that downloading to D ever used to be any quicker? Or do you just feel that the bug's always been there? Y'see I don't think its a bug as such - I just think that the slower write speed is limiting the download rate. I know that the d/l rate is actually slower than the average write speed of CF, but I don't think its necessarily that simple - writing to CF goes via a whole different route than C, so there's more than just the CF write speed at issue.

Feel free to contact Psion about this one, but I think we're unlikely to get a response. I'll bet the response will be along the lines of 'be thankful that it works at all!'

Mebbe I'm just pessimistic, mebbe you'll get more joy, I dunno...

[dukibean]

The write speed to CF using EPOC is pretty dire. Put your CF card into a PC card adapter, use it in a laptop and see how well even a fairly crud card performs. Only today did I copy the 25mb PsiWin exe to a crappy Expansys branded 128Mb in well under a minute. EPOC would take the rest of this afternoon to do the same job. . . Bit of a shame, really.

[tomas]

Thanks much for bug confirmation.

I think we can report it oficially to the teklogix now -

Hallo, teklogix, are you here ? We have a serious bug to report :-)

Thanks.

[Davidish]

Hmmm... now I might be wildly wrong here, but reading all your posts made me wonder this:

Has anybody tried copying a big file direct from C: to D: and measuring the time that takes? HOw does it compare to the opera and nftp rates? THat might suggest it is not to do with the d/l process at all.

But perhaps it is not a CF write issue, or an OS bug either that you've all found, but rather a fundamental 'feature' of the Netbook motherboard that the bandwidth from the main memory to the CF port is so narrow that you can only achieve these slow write rates to CF on the compact flash card - regardless of the CF card's actual write speed.

after all the data has to be transfered from C: to D: SOMEHOW and this on-board bus might be the problem. Which would mean there is no solution for speeding things up short of a new motherboard design :(

of course I might be completely wrong and would be happy to see evidence to the contrary. But I have a horrible feeling (based on my limited tests with my own (admittedly slow) CF card) that I'm not.

D.

[Davidish]

PS: Actually, having said all that, I am not ruling out it being an OS problem coz I guess that could easily be responsible for the slow down too. SO, I suppose contacting psion would still be a very good idea - if someone knows who to contact to clear this up.

[tomas]

Originally posted by Davidish

Has anybody tried copying a big file direct from C: to D: and measuring the time that takes?



Of course, as I wrote before. By copying large file from C to D I measured approx. 190 kB/s write speed of my CF. Far far away from download speed of 10-30 kB/s.

All i/o CF operations are OK, ecxcept of download from the net, so it cannot be netbook HW fault or media speed problem.

[Davidish]

Sorry tomas. I missed that post whilst scrolling through the thread. I guess I AM wrong :p

[tomas]

Originally posted by diem
Tomas, you mentioned that the numbers have changed between your versions of nFTP - what are your version numbers and has it sped up or slowed down?


With old OS I used last standalone nFTP version (1.30).

It downloaded to the C and D with the same speeds - some 80 kB/s, but never more than that - so there was probably some sw limit. BUT - as i said - to the C and D with the same speed ! It is why I thought that the download slowdowns to D were only Opera's problem, not OS.

Now, with the new OS, there is ibuilt nFTP version 1.31.

And this version performs the same as opera did. The old nFTP "limit" of 80 kB/s was broken and I can download with more than 150 kB/s, BUT ONLY TO C. Download to D with this new version is the same as measured with opera - only some 10-30 kB/s.

So, the anwer is, the new version has speed up (C) A N D slow down (D). The problem has to be in the OS.

[diem]

I get what you've been trying to say now Tomas, and I can see why you're annoyed about this!

I'm with you - I'd like to hear what Psion (Harvey?) has to say about this...

[harveyr]

Wasn't aware of this issue until now. I don't know the reason for why you are experiencing such disappointing download speeds to the D drive but will endeavour to find out for you....

Harvey

[allan]

Only noticed this thread a month after it was active, but I've noticed this behaviour too. It's always been there - well it was with opera 3.x at any rate. I emailed the developer about it, and although I don't think I got a reply, Opera 5 seems slightly improved (Opera 3 was almost unusable when downloading to CF card).

Copying files directly from C to D or even from E to D is relatively fast, so it is not a property of the CF card itself.

I suspect an issue in how the data is being transferred - eg perhaps Opera is writing byte by byte, while a file copy is more likely to do some kind of block transfer; and maybe the byte copy routines take a much larger hit from the lower transfer speed of CF.

Graham