[c-nsp] OSPF Convergence time on Dynamips

[Christopher.Marget at usc-bt.com]

> > I got a huge performance boost when I moved the dynamips directory to
> > a reasonably fast disk subsystem.
> 
> uh, why was this?  Surely you weren't paging your inbound packets to disk
> before forwarding them?  Or was this purely for startup / shutdown?

I never investigated it very far, but I think dynamips puts router memory onto disk.  Maintaining the myriad timers in a large OSPF LSA database might be a lot of memory i/o?

I also happened to be using a particularly low-performing disk subsystem*, so maybe I ran into something that doesn't matter in most implementations.

When I changed the dynamips filesystem it quadrupled the size of the topology I could run.

/chris

* ext3 filesystem on a virtual (xen) block device in a file in a zfs filesystem on a raidz2 disk pool.