netreg Revisited
Jason Antman
jantman at oit.rutgers.edu
Tue Mar 29 12:41:00 UTC 2011
John Wobus wrote:
> We rolled our own system years ago that's pretty
> much as you describe netreg and we have never
> changed from restarting dhcpd to using OMAPI. I was
> influenced by an on-list dhcpd developer comment that
> they'd like to replace OMAPI, and also I was
> uncomfortable with host configs in the lease file
> rather than the config file. Also, by regenerating
> the entire config file, we easily eliminate the
> possibility of dhcpd and our database getting out
> of synch. [...]
I distinctly remember the developer comments somewhere about OMAPI also.
1) I also was uncomfortable with what, at least at the time, seemed to
be a lack of real-world working examples of doing things with OMAPI...
i.e. working libraries (specifically PHP) that are vouched for by people
who use them every day, along with sample code, etc.
2) I was concerned about not having an authoritative data source. We
need an authoritative source of IP/MAC assignments directly in a
database, where we can manage access control (to the lease information)
for many hosts and applications. I don't see a way to do that with the
leases file, and I remember reading at least a few comments about OMAPI
not being 100% reliable or atomic.
3) Before our overhaul, we'd been using the Masney LDAP patch on *old*
dhcpd code. We wanted to move away from LDAP to MySQL, for a number of
reasons - one being extremely poor performance. When we had to restart
dhcpd (subnet changes or something like that), it was taking over 1
minute to start. In my investigations (dhcpd serving 50k+ possible
addresses), I found that we had a ~20-30MB leases file, and that was the
root cause of the long startup time. As a result, when we restart, we
also parse the entire leases file into MySQL and regenerate it with only
current leases, or leases expired in the last LeaseTime*2.
If anyone has an actual write-up on how they integrated OMAPI, the
reliability, performance, etc. I'd be very interested in reading it.
Jason Antman
Rutgers University
More information about the dhcp-users
mailing list