[pptp-server] Maximum number of users

[Torge Szczepanek]

Am Mon, 2002-04-15 um 07.35 schrieb Neale Banks:

> True, but you forgot the "please also fix the algorithm" bit.

Yes, but there is currently no patch that I know about to fix this
problem and which was tested. I don't wanna experiment on our production
system with own kernel patches or untested patches.

There is a hashed device lookup patch on the archive of the linux kernel
mailinglist:

http://www.uwsg.iu.edu/hypermail/linux/kernel/0101.0/1215.html

I never tried this one, because I never touched the limits of our
hardware.

> I wonder if anyone has ever estimated the hit of this function in
> allocating a name for say the 2001th ppp device (noting that if you're
> doing that, you've possibly got a significant proportion of those 2000 ppp
> devices doing encryption and/or compression - a non-trivial load of
> itself)?

I think that 2000 ppp connections would be a real problem. Using the
newest hardware (biggest AMD or Pentium IV CPU) I would guess that you
can reach about 1000 users in our user-enviroment with one single
server. Our users all have a 100MBit connection to our server but most
of them do not transfer much data all the time. In other ennviroments
you could end up with a much lower number of users for one machine, when
users transfer much data all the time.

> Of course, a patch to fix the underlying deficiency would be far superior
> to this mere diagnostic print.

Yes! I agree to that! This is one of my top wishes for the linux kernel!

-- 
Torge Szczepanek