Operating System - HP-UX
1832235 Members
2861 Online
110041 Solutions
New Discussion

Re: Source quench in hpux 11.00

 
SOLVED
Go to solution
Achilles_2
Regular Advisor

Source quench in hpux 11.00

Dear all,

I am several hp A-class servers that running 11.00. They are on the same routed network and on same LAN media type: FDDI.

I find that the traffic initiated from Cisco Catalyst switch to HP UX server come across failure rate, depending on the size of packet, bigger size suffered more. Packet with minimum size likes 100 byte was fine. For a typical 1000 byte packet, almost 30 to 40 percent packet lost.

So, I try to set
#ndd -set /dev/tcp tcp_mss_max 1460
#ndd -c

As a result, I am performing a ping from the HP-UX servers and received a source quench from the pinged system.

Setting the ndd parameter ip_send_source_quench to 0 can be an effective way to deal with the messages.

I know that the source quench is a way for the pinged system to tell the pinging system that it is sending ICMP traffic too fast for it to handle i.e. giving a "please slow down" signal.

Source Quench is sent back from a host when it cannot deliver packets to the destination socket quickly enough. There is a bug in HP-UX 11.0 that causes it to return excessive SQ messages as well.

So, is there a patch to fix this problem in 11.0?

Do anyone give me some advises, thanks.

Alex
3 REPLIES 3
steven Burgess_2
Honored Contributor
Solution

Re: Source quench in hpux 11.00

Hello

Have a look at PHSS_29963

HTH

Steve
take your time and think things through
Bharat Katkar
Honored Contributor

Re: Source quench in hpux 11.00

Hi
You can monitor source quench by
# netstat -p icmp
Regards,
You need to know a lot to actually know how little you know
rick jones
Honored Contributor

Re: Source quench in hpux 11.00

11.00 IP code was entirely too eager to send source quench messages. You might as well disable them via ndd.

As for the packet losses, I have a hard time seeing how altering tcp_mss_max would have helped - unless perhaps the FDDI interface(s) in your HP systems were losing races in the FIFOs between the network and the host memory. I would check the lanadmin statistics for the NICs and see if there were any FIFO settings for how soon the NIC might start to DMA packets into the host.

BTW, what sort of system and FDDI NIC is this? If this is the old HP-PB FDDI NIC, in something like a K class, the best you could hope for in that would be something along the lines of 65 Mbit/s.
there is no rest for the wicked yet the virtuous have no pillows