m�n 2003-03-03 klockan 16.34 skrev Sheahan, John (PCLN-NW):

> If I am calculating this correctly, since each squid server keeps track of
> it's incoming connections from the end users on a port basis, the maximum
> amount of concurrent connections for each squid server would be limited to
> the amount of available TCP ports. I am calculating this as 64535-1024 =
> 63511 maximum total connections.
> Can anyone please verify this for me? Thanks in advance.

The actual limit is much lower as you also have to account for TIME_WAIT
blocking recently used ports.

In theory much higher limits can be acheived by using smarter port
allocation schemes than available in stock TCP/IP kernels. The
theoretical limit is only per source/destination IP:port pair, but most
TCP/IP kernels take the safe path and denies reuse of the source port
while there is other uses active (or in TIME_WAIT) even if there is
differences in the tuple sufficient to distinguish the connections. 

However, in practice your run into many other limitations long before
the above limit is reached.

  - number of sockets/filehandles available in the system
  - number of filedescriptors available to a single process
  - CPU usage
  - Interrupt management

And Squid (and most OS:es) is certainly not designed for such wast
amount of concurrent connections.


-- 
Henrik Nordstrom <[EMAIL PROTECTED]>
MARA Systems AB, Sweden

Reply via email to