VOGONS


First post, by Grzyb

User metadata
Rank l33t
Rank
l33t

It's well-known that at least some of the packet drivers for 3Com PCI cards are buggy.
Now, let's try Intel...

NIC: Intel PRO/100 VE
CPU: Pentium III 800EB
HDD: WDC WD1200JB, running in ATA 100
software: FTP.EXE from mTCP Jan 10 2025, MTU 1500
drivers: see the attachment

all results in KB/s

native PD:

download to NUL: 8877
download to HDD: 193 !!!

NDIS+DIS_PKT:

download to NUL: 9042
download to HDD: 4321

ODI+ODIPKT:

download to NUL: 9042
download to HDD: 4264

So, when you encounter mysterious problems with some PD software, you know what to do...
And if you find a better PD for PRO/100 VE, let me know!

In 2003, I voted in favour of joining the European Union. However, due to recent developments - especially the restrictions on cash usage - I'm hereby withdrawing my support. DOWN WITH THE EU!

Reply 2 of 10, by Grzyb

User metadata
Rank l33t
Rank
l33t
Matth79 wrote on 2026-06-10, 17:13:

Not sure if the Seth Simon from here would be any better

Only marginally better...

download to NUL: 9042
download to HDD: 214

In 2003, I voted in favour of joining the European Union. However, due to recent developments - especially the restrictions on cash usage - I'm hereby withdrawing my support. DOWN WITH THE EU!

Reply 3 of 10, by MagefromAntares

User metadata
Rank Member
Rank
Member
Grzyb wrote on 2026-06-10, 15:29:

native PD:

download to NUL: 8877
download to HDD: 193 !!!

While I haven't yet debugged the .com file, if there is a so large difference between downloading to NUL and to the HDD it usually means that there are no/little buffering of writes to the disk, which means that the HDD has to write each byte or small group of bytes separately, needing to seek the proper track and position again.

EDIT: If you wish to check if this is the issue, MS-DOS 6.22 comes with RAMDRIVE.SYS, allocate a disk in memory to try, if this is the case that will significantly increase the performance.

"A process cannot be understood by stopping it. Understanding must move with the flow of the process, must join it and flow with it." - Dune

Reply 4 of 10, by Grzyb

User metadata
Rank l33t
Rank
l33t

Currently using the PD by Seth Simon.
Download to RAMdisk is fine - same speed as to NUL.

BTW: RAMDRIVE.SYS is limited to 32 MB - what's the recommended replacement?
And why does "ramdrive.sys /e" in MS-DOS 7.10 require "device=c:\windows\himem.sys" in CONFIG.SYS? In this DOS, HIMEM.SYS is loaded automatically!

In 2003, I voted in favour of joining the European Union. However, due to recent developments - especially the restrictions on cash usage - I'm hereby withdrawing my support. DOWN WITH THE EU!

Reply 6 of 10, by Grzyb

User metadata
Rank l33t
Rank
l33t
Joseph_Joestar wrote on 2026-06-11, 10:59:

Try XMSDSK.

Thanks, seems to work fine!

Meanwhile, testing with another client - FTP2P...

download to RAMdisk: 11224
download to HDD: 217

In 2003, I voted in favour of joining the European Union. However, due to recent developments - especially the restrictions on cash usage - I'm hereby withdrawing my support. DOWN WITH THE EU!

Reply 7 of 10, by wierd_w

User metadata
Rank Oldbie
Rank
Oldbie

Unrelated to packet driver testing, but just random trivia:

Xmsdsk hosted ramdisks can be used with the drvspace driver. You just need to have the drvspace.000 file present on the ramdisk, then point scandisk at it with the undocumented /mount argument. Scandisk will mount it for you silently.

1gb is the max officially supported size though.
if you have 64kb of upper memory to just blow (drvspace.sys /move in config.sys will relocate it to umb instead of dos memory), it can be useful. A whole volume can be ftp'd then mounted this way.

Derail over--
Carry on!

Reply 8 of 10, by EduBat

User metadata
Rank Member
Rank
Member

I had problems with ne2000 packet drivers on my 486, which disappeared when I disabled both the processor and the external caches. It's weird, I don't know why it worked...

Reply 9 of 10, by MagefromAntares

User metadata
Rank Member
Rank
Member
Grzyb wrote on 2026-06-11, 12:49:
Thanks, seems to work fine! […]
Show full quote
Joseph_Joestar wrote on 2026-06-11, 10:59:

Try XMSDSK.

Thanks, seems to work fine!

Meanwhile, testing with another client - FTP2P...

download to RAMdisk: 11224
download to HDD: 217

Hmm it seems that FTP2P is another client that doesn't do proper buffering of data before writing to disk.

As an alternative to RAMDRIVE.SYS or XMSDSK you might also try it with SMARTDRV with write-behind cache enabled: SMARTDRV <drive letter>+ <other parameters...>

The "+" after the drive letter tells SMARTDRV to enable write-behind caching of the given drive, so unless if the given client bypasses DOS File Write interrupts/functions it will be cached by SMARTDRV instead of directly written to the disk, it also has the advantage that you don't have to copy the file from the RAM disk to permanent storage after transfer, however there is a big disadvantage to this method, whenever shutting down the machine you need to run SMARTDRV /C to make sure that the caches are flushed so data doesn't get stuck half written(And crashes are also more likely to result in lost data).

"A process cannot be understood by stopping it. Understanding must move with the flow of the process, must join it and flow with it." - Dune

Reply 10 of 10, by Grzyb

User metadata
Rank l33t
Rank
l33t

Using the PD by Seth Simon, and mTCP FTP...

smartdrv c+

download to HDD: 286

Seriously, can't suspect the FTP clients - both work fine with both NDIS and ODI, even without SMARTDRV.
There's something wrong with the Packet Drivers, but I can't even imagine what exactly...

In 2003, I voted in favour of joining the European Union. However, due to recent developments - especially the restrictions on cash usage - I'm hereby withdrawing my support. DOWN WITH THE EU!