[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

asymmetric transfer rates



Hi guys,

I was playing with my micro home net recently and noticed a weird
phenomenon. Transfer rates differ drastically in different directions.
Configuration: 
1) OpenBSD-current as of Aug-16 on Sparc LX.
   SBus card containing both AMD Lance and "esp" scsi
2) OpenBSD-current as of June on i386 (Celeron) with Netgear card
   pn0 at pci0 dev 11 function 0 "Lite-On PNIC" rev 0x21: irq 9
3) Crossover cable between them.
4) ip filtering disabled.
5) Both computers idle.

Now ftp'ing big empty file (dd if=/dev/null of=empty seek=10000 bs=1k).
from sparc to i386:
ftp> get empty /dev/null
local: /dev/null remote: empty
227 Entering Passive Mode (x,y,z,2,93,240)
150 Opening BINARY mode data connection for 'empty' (10240000 bytes).
226 Transfer complete.
10240000 bytes received in 336.88 seconds (29.68 KB/s)

from i386 to sparc:
ftp> get empty /dev/null
local: /dev/null remote: empty
227 Entering Passive Mode (x,y,z,1,159,134)
150 Opening BINARY mode data connection for 'empty' (10240000 bytes).
226 Transfer complete.
10240000 bytes received in 9.13 seconds (1.07 MB/s)

Yeah, I know sparc LX is no match to Celeron performance, but 36 times
difference between network receive and transfer rates?!

Another thing: collision rates are quite high on these interfaces (grow
during transfers either way). Why collisions on crossover cable? Do I
need to force cards into full duplex (if it is possible at all on 10Mb?)
to avoid collisions?

Thanks
Greg