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

Re: routeing table size gotcha (IPv6 ?)

>I am assuming that this is the result of the IPv6 integration, but tell me
>otherwise. Having innocently tried to upgrade our router kernel to
>2.6-current (to support the if_dc driver for performance) I suddenly found
>that the kernel was running out of memory.
>Where in the past, carrying 72,000 odd routes (145,000 entries) used about
>11M of routetbl entries (150 byes or so each), with the new kernel I got to
>60,000 routes (120,000 routetbl entries) for 16M of space (about 260 bytes
>an entry).
>With the growth of addresses from 4 to 16 bytes, is this major size increase
>due to (a) alignment issues, (b) prefix/netmask/etc. additions or (c) me not
>knowing what is going on ?
>If there is someway to optimise the FIB for full-Internet route carrying
>routers, I think that this would be well received.

	I assume you are carrying bunch of IPv4 routes, not IPv6 routes.

	I don't think IPv6 integration affected your problem.  IPv4 routing
	entries are kept unchanged.  sys/net/route.c handles address families
	separately, and it does not require change in size even if we integrate
	IPv6 into the kernel.
	Just in case, what happens if you remove "options INET6" from your
	kernel configuration?  It would decrease kernel code size and
	may make things better for you.