TCP time out sometime on HPUX 11

[Romeo_Lou]

Dear,
There are two HPUX in server room, but one of HPUX (Server B)encounter TCP time out sometime when user ping from OA seg, but ping from OA seg to another server(server A) is ok.
And ping from server A to server B is ok too.
Ps server A and B within the same seg different with OA seg.
Does there any parameter to fix this issue on HPUX?
or any suggestion?
Tks

[Scott Van Kalken]

you can use ndd to check and or set such parameters


the man page is the most comprehensive guide to ndd.

man ndd

name to get/set ? ndd -get /dev/tcp
value ?
length ?
7200000


two hours by default.

If this is a firewalled environment and you're using checkpoint firewall, yo can get into a situation where by default the HP is a two hour interval and checkpoint is a one hour interval (by default).

Scott.

[Romeo_Lou]

Dear,
all of servers and users are in intranet, so that should no firewall issue, right?,
and does there any parameter about tcp time out , E.g. for deatil, maybe within 1000ms the user can get reply from server, but get timeout when greater than 1000ms.

tks

[Scott Van Kalken]

yup
ndd can get and check these parameters, by default it's 2 hours.

[Romeo_Lou]

sorry, which parameter, there are mayn parameter when use "ndd -get /dev/tcp ?"

[Bill Hassell]

This likely a network setup problem. Verify that all your 100BaseT LAN cards have valid connections (ie, 100 Mbit, full duplex if this is correct). If you are expecting 100 full duplex but see 100 half duplex, then autonegotiation has likely failed.

Autonegotiation went through growing pains for 100BaseT and there are basically three ways to determine the speed/duplex settings (IEEE 802.3u). Before 1994, the N-way negotiation standard did not exist so Auto-sense (10 Mbit) and a parallel detection method must also be used. This involves Fast Link Pulses and a lot of details to make it all work.

However, successful auto-negotiation requires agreement on the methods and timing, plus LAN cables that are less than 35 meters or longer than 41 meters. Because auto-negotiation uses special pulses and detection windows, certain cable lengths may fail to negotiate and the standard says: if auto-negotiation fails, default to 1/2 duplex.

However, if the LAN card is at 1/2 duplex and the switch is at full, the result is a huge number of errors and in some switches, the port may be automatically disabled.

So unless your machines are in a testing lab where short cables and switches are swapped on a daily basis, it is simpler to select 100 full-duplex for BOTH sides. Note: if you set the computer side to 100FD but leave the switch to autonegotiate, the switch may fallback to 100HD which is no good. If the switch cannot be changed, try a much longer LAN cable to see if autonegotiation can be successful with a longer (or shorter) cable. As mentioned, the 'gray area' is 35-41 meters. And this can explain why two different machines negotiate differently--the LAN cables are different lengths.