Re: performance (was Re: cache not up to date)

> >Any tips on Digital UNIX kernel parameters ? anyone ?
>
> Well, my experience has been that fixing kernel bugs is much
> more important than tuning :-). If you are running Digital

Kernel bugs are really tricky business. Especially
for users like me, it is difficult to know whether if
it is the kernel fault or my fault. :(

> Unix 3.2C, be sure to get patch OSF350-146. This fixes a TCP
> window size bug (which causes the machine to run out of mbufs),
> plus lets you fiddle with the socket listen queue sizes.

> You may also want to look for a patch which employs an mbuf
> clustering algorithm (sorry, don't have a patch number).

I will check with the DEC people locally. Thanks for the
tips.

By the way, it come to my interests about some statements in
the man page of listen() that said that ...

---Quoted from man listen---
The maximum queue length (backlog) that the listen() function can specify
is eight. The maximum queue length is indicated by the SOMAXCONN value
in the sys/socket.h header file.

---
Now, if Digital UNIX library limits the maximum queue length.
we probably going to have problems even the kernel somaxconn is
set pretty high and the listening queue for Harvest is set 
to a reasonable value.
Okay... I use Harvest Cache V2. I assume it should behave
similar to Squid.
*8)
Ong Beng Hui
ongbh@singnet.com.sg
...Yet Another Day in a ISP Business
...and they lived happily ever after