UDP send/recieve buffer size

[Md. Minhaz Khan]

Hi,

If some one was to suggest, "increase your kernal tunables related to UDP, in particular the UDP send/recieve buffer size".... then what would they mean?

How can I find out what this current value is?
I'm using HP_UX 11i B.11.11.


Thanks
Minhaz

[Steven E. Protter]

Shalom,

This answer is OS dependent.

for 11.23 and 11.31
kctune -l

For 11.11
kmtune -l

Shows all kernel tunables.

BUT!

I think this is really a network tuneable, which means its all in /etc/config.d/nddconf

The ndd utility allows you to display and change these mostly on the fly.

SEP

[Md. Minhaz Khan]

Any help from others expert??

Thanks
Minhaz

[James R. Ferguson · Accepted solution]

Hi Minhaz:

You might do:

# ndd -h supported

...and from those results:

# ndd -h socket_udp_rcvbuf_default

# ndd -get /dev/sockets socket_udp_rcvbuf_default

Regards!

...JRF...

[Md. Minhaz Khan]

Thanks to all, I have another quey
if i want to set value "socket_udp_rcvbuf_default" to 2147483647

then what is the command??


Thanks
Minhaz

[James R. Ferguson]

Hi (again) Minhaz:

To set the value do:

# ndd -set /dev/sockets socket_udp_rcvbuf_default

...where the range of values for is seen from:

# ndd -h socket_udp_rcvbuf_default

Regards!

...JRF...

[Steven E. Protter]

Shalom again:

ndd -set /dev/sockets socket_udp_rcvbuf_default 2147483647

To make it permanent to survive a reboot you will need to record the change in /etc/rc.config.d/nddconf

SEP

[rick jones]

Someone else has already pointed-out that while TCP settings are made to /dev/tcp and IP settings to /dev/ip UDP settings get made to /dev/sockets... so much for foolish consistency in the HP-UX networking stack :)

To what end is someone asking you to make the default UDP socket buffer receive size so very large? At first blush it appears to be a MASSIVE KLUDGE for an application with a desparate need for end-to-end flow control that does not have such.

Also, it seems a lame practice to set the *default* so high. *Perhaps* set the max that high, but not the default. Then the application can set its own sockets, and only its own sockets to such rediculously large values with a setsockopt(SO_RCVBUF) call. Unless, of course, said application is also so terribly broken it does not have a way to set socket buffer sizes itself.