On Thu, Aug 31, 2006 at 07:44:30PM +0200, Per Engelbrecht wrote: > Hi Henning, Claudio and everybody else on the list, > > I've been working with kern.maxclusters up and down between 6144 and 32768. > Whenever I raise the value I get a short lived drop in "Kbytes allocated > to network' (netstat -m) only to see a clime again within minutes. > Average is 75-90% > Anything above 70% result in a '* * *' when doing a traceroute from > whatever lookingglass around the world. > > For info: this BGP router runs a very simple pf setup, but disabling pf > does not change anything at all(!). > Additional info: net.inet.ip.ifq.maxlen (>50 but <250) does not change > anything neither. > > In short; I'm out of great ideas and I have a whole lot of customers > (and management@) shouting louder and louder. > If I can provide any extra info/documentation besides what have been > given i previous mails, then please let me know. I'm about to enter the > "red zone". Need help. >
If that is the box with 4GB of RAM take about 2-3GB out of the box and retry. The problem is that the kernel memory space is limited and a lot is wasted for stuff that you don't need on a router (like about 200MB for the buffer cache). Sometimes more memory is not better. -- :wq Claudio
