I've done some more measurements, and have found a
bit of a mystery (to me, at least):
________ ________
| | |<--89 Mbps(ftp)-------------------->| | |
| | | ___________ | | |
| | |<--89 Mbps(ftp)-->|e| |e|<-57Mb>| | |
| |e| |t| |t| (ftp) |e| |
| |t| |h| |h| |t| |
| |h| |0|_____|1| |h| |
| |0|<--17 Mbps(ftp)---|-|kernl|-|------>|0| |
| | |<--83 Mbps(ttcp)--|-|bridg|-|------>| | |
| | | | |_____| | | | |
| | |<--16 Mbps(ftp)---|-|user |-|------>| | |
|____|_|<--83 Mbps(ttcp)--|-|bridg|-|------>|_|____|
Machine A | |_____| | Machine C
|_|_____|_|
Machine B
(bridge)
Okay, so, hoping that this diagram is intelligible,
why does TTCP show such drastically better performance
than FTP, but /only/ when it is done over the bridge?
Conversely, why does FTP perform so poorly, but /only/
when it is done over the bridge? As a note, SCP
exhibits the exact same behavior: good direct in any
permutation, very poor over either kernel or userspace
bridge.
Notes:
- A is a 1.8Ghz Pentium with a 3C905.
- B is a 500Mhz Pentium with a 3C509B and an epic100.
- C is a 500Mhz Pentium with a 3C509B.
- ttcp was run with '-n 20480 -s' (tcp packets).
- The userspace bridge is just a simple program that
listens to both interfaces in promiscuous mode and
copies packets from one side to the other.
Any ideas?
Thanks,
Alen
__________________________________________________
Do You Yahoo!?
Great stuff seeking new owners in Yahoo! Auctions!
http://auctions.yahoo.com
_______________________________________________
Bridge mailing list
[EMAIL PROTECTED]
http://www.math.leidenuniv.nl/mailman/listinfo/bridge
