[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Very huge email service



On Thu, 25 Nov 1999, Chris Wagner wrote:
>At 04:42 PM 11/24/99 -0700, Ken Gunderson wrote:
>>Speaking of mondo email systems, I was talking with a guy today who
>>mentioned that their server would max out at 64K users.  It was a Linux
>
>I believe that is the limit of the UID space.  I assume you can hack
>something to increase it arbitrarily large.  But that would involve
>increasing the address space, which is not as simple as changing a flag
>somewhere and recompiling.  AKAIK, the address space is 16 bits which gives
>you up to 65536 UID's (plus root).  The "default" UID space on Debian is
>from 1000 to 29999.

Can't 64bit architectures already do 32bit UID's?

The problem with >32767 on intel of course is that some programs make it a
signed short and check for >0...

>So for VHES, it would be best to just bypass UID's altogether and use
>another identification scheme.

Probably qpop with all accounts having the same UID.

Also I think that /etc/passwd can be more than 64K entries if you have a
database of the data.  If not then performance becomes ridiculously low when
you get 32K entries.
I am working on an AIX system with 27232 accounts in /etc/passwd.  Before we
ran "mkpasswd -f" to create the databases the load average was always >30. 
Idle time was never anything other than 0%, and the machine was slow as a dog.
Now the load average is rarely higher than 10, and sometimes below 1, we
sometimes see idle time, and performance is often good.
This is with users who don't do that much email, and there are two PPC CPUs
to parse /etc/passwd.
Having an Intel machine with 64K /etc/passwd entries and no database would be
much worse than this.

-- 
Electronic information tampers with your soul.


Reply to: