tjh+@andrew.cmu.edu (Tom Holodnik) (02/04/89)
We seem to be running into an interesting problem here with Annexes and large routing tables published by RIP. Recently, we allowed the NSFnet routes to be published onto our campus backbone. The routing tables contain approximately 312 routes, taking up about as many mbufs in memory on the Annex. Here's the output from "netstat -m:" annex6: net -m 500/599 mbufs in use: 45 mbufs allocated to data 8 mbufs allocated to packet headers 15 mbufs allocated to socket structures 26 mbufs allocated to protocol control blocks 312 mbufs allocated to routing table entries 20 mbufs allocated to socket name 2 mbufs allocated to interface address 8 mbufs allocated to process context 64 mbufs allocated to incoming network i/f packets *** 2 mbufs missing *** 149 Kbytes allocated to network (83% in use) 0 requests for memory denied I'm not sure what the upper limit on the number of mbufs is on the Annex, but my guess is that 599 must be near the limit. What made us notice this is that there are occasional instances where users receive messages to the effect that there are no available buffers in memory for new connections. I'm not sure that there is much that the developers can do about this, apart from adding more memory to the system (as a daughter board, perhaps?), if I'm correct. Does anyone know more about this, that can verify that this is what's happenning? Thanks, Tom Holodnik
loverso@Xylogics.COM (John Robert LoVerso) (02/10/89)
In an article in comp.sys.encore, Tom Holodnik <tjh+@andrew.cmu.edu> writes: > We seem to be running into an interesting problem here with Annexes and > large routing tables published by RIP. Recently, we allowed the NSFnet > routes to be published onto our campus backbone. The routing tables contain > approximately 312 routes, taking up about as many mbufs in memory on the > Annex. You're correct. When an Annex boots, it allocates a static pool of mbufs, currently 600 mbufs per 1Mb, where each Annex mbuf is 256 bytes (double that of BSD). The number of mbufs is never grown or shrunk. For an Annex-I, this means you will be tight on mbufs if a local gateway is RIP'ing routes equivalent in number to the Internet routing tables. In Annex R4.1, you can disable the Annex RIP-listener (its routed process) and instead rely upon a default route to a local gateway that will issue ICMP redirects as needed. Its possible that there'll be a new per-annex EEPROM parameter for configuring the number of mbufs allocated at boot time, but thats not definite yet. R4.1 is still in beta test, and will be available in a few weeks. -- John Robert LoVerso, Annex Software Engineer Xylogics, Inc. [formerly with Encore Computer Corp] encore!xylogics!loverso, loverso@Encore.COM