Concept Review: Cross-over versus Hub

[snorlaxx]

Sorry for this super noob question but...

We had just concluded testing of FTP and TFTP transfers between two PCs using cross-over cable and LAN. TFTP as expected is faster than FTP. For TFTP, cross-over was faster than with a hub. But in FTP, the Hub was faster than the cross-over. We were wondering how this happened. Shouldn't be the cross-over cable have faster transfer compared to the hub. The hub is capable of 10/100 Base transfers. TFTP data was 1MB, while FTP data was 8MB.

Generally speaking, i would like to ask if the Cross-over cable is a half duplex or full duplex (or is this defined elsewhere?)? and should the premise always be Cross-Over speed > Hub speed.

Thankyou in advance!

[happybunny]

i have no idea.

My guess is that when plugged into the hub, and your machines are set to auto as they should be in a hub, they negotiate 100/full.

When using cross-over cable, maybe they only negotiate 10/half 100/half.

What OS? When you connect either way, do a ethtool eth0 on both to see what the NIC is set at.

[snorlaxx]

Originally posted by happybunny
What OS? When you connect either way, do a ethtool eth0 on both to see what the NIC is set at.
1 PC runs 2.4.18 kernel on RH7.2. The other is an SH custom board using a bsp patched 2.4.18 kernel.
maybe if I try using ethereal, i might get answers. thanks!

[dysharmonic]

As far as I know, a hub, as opposed to a switch, uses the CDMA/CS method in order to let each node that is sharing a single segment of the network w/ the other pcs, send a packet. And in this situation it's always half duplex.

If both NICs are directly connected to one another, then I guess as long as them both could support full duplex, then they will likely switch their operation to that mode...

Well, maybe I'm wrong...I'd like to know too... :D

[thaddaeus]

in crossover they will match the speed of the other if possible, same goes for the hub, but the nics will match hub speeds and the hub will match the nics. The cross >= hub defanantly, especially with multiple machines attached(hub), now if your using a router then the speeds should actually be very close, but instead of the whole network falling to 10 because of 1 10 nic, only that machine(and any machine connected to it) will suffer bad transfer rates.
The whole TFTP FTP thing is interesting though, did you do multiple tests to determine whether or not it was just a fluke, the file size for the transfers is a bit odd too 8mb for ftp and 1mb for tftp, try using same file for bothe transfers.

[snorlaxx]

The time difference between my LAN FTP and Crossover cable FTP is 0.2secs. Is this a big difference? or could this be negligible? My hub had only both PCs connected at the time of measurement. So i guess the timings should fairly be the same.

TIA!

[snorlaxx]

Hey guys!

Why and when does FTP become faster than TFTP?!?! We have this custom board that we ran on Emb. Linux using busybox. After testing, we saw that TFTP was slower than FTP. We performed the same testing with an x86 board running emb. Linux as well. And TFTP was indeed faster. Our TFTP server is a Windows application. We couldn't help but compare the result of the custom board with the P3 i386 PC.
I tried to blame Windows for performing additional protocols besides tftp, but since the same process was made with both boards, i end up with nothing credible.

Need your help!
TIA!

[Leigh]

You have more than one fastor at play here.
As a rule of thumb, a hub will be half-duplex.
A crossover will probably be full-duplex.
If you are using a switch, and not a hub, it could be either.

The difference becomes noticable on FTP before it does on TFTP.
TFTP is UDP, FTP is TCP. UDP sends a packet and doesn't care if it arrives or not. TCP sends a packet and waits for an acknowledgement. Therefore a half-duplex link will be slower for TCP than UDP. (and therefore FTP slower than TFTP.)
However, on Full duplex, both hosts can talk at the same time, so the delay whilst waiting for an ACK is less.

So, on a hub, which will be half-duplex, you can expect FTP to be slower. Not the case for TFTP.

0.2 secs difference for an 8MB transfer? Negligible. And probably not repeatable.

As far as the comment about hubs using CSMA/CD, ALL ethernet devices use CSMA/CD, unless they use CSMA/CA, which is these days reserved only for Appletalk and 802.11.