rees@pisa.ifs.umich.edu (Jim Rees) (01/17/91)
In article <79764@unix.cis.pitt.edu>, csfst1@unix.cis.pitt.edu (Charles S. Fuller) writes:
The SR10.1 version of routed was afflicted with a form of Alzheimer's;
it would dutifully gather information from other routed's, then forget
what it had heard. There is a patch available, which (I believe) is
essentially the SR10.2 routed. Contact the Apollo Response Center.
The sr10.2 routed has this same problem. Either get sr10.3 or the routed
patch, number 108. sr10.1 tcp had other problems as well, you should at
least get sr10.2.
But since he can't even ping other hosts on his subnet, this isn't just a
routing problem. I too would be worried about the address given by netstat
-i. The first thing I would do in this case is install sr10.2 or sr10.3.spear_r@apollo.HP.COM (Robert Spear) (01/17/91)
In article <4f3e1cec.1bc5b@pisa.ifs.umich.edu> rees@citi.umich.edu (Jim Rees) writes: >In article <79764@unix.cis.pitt.edu>, csfst1@unix.cis.pitt.edu (Charles S. Fuller) writes: > > The SR10.1 version of routed was afflicted with a form of Alzheimer's; > it would dutifully gather information from other routed's, then forget > what it had heard. There is a patch available, which (I believe) is > essentially the SR10.2 routed. Contact the Apollo Response Center. > >The sr10.2 routed has this same problem. Either get sr10.3 or the routed >patch, number 108. sr10.1 tcp had other problems as well, you should at >least get sr10.2. > >But since he can't even ping other hosts on his subnet, this isn't just a >routing problem. I too would be worried about the address given by netstat >-i. The first thing I would do in this case is install sr10.2 or sr10.3. Jim, the sr10.2 version of routed is not "broken" as in 10.1. 10.1 routed was prematurely aging routes (metric 16) and therefore, you would loose the route to the destination. 10.2's routed would cause an access violation if invoked as /etc/routed -t (with no filename). It worked fine in every other way. The recommended patch for 10.2 only is patch m0183 and m0139 (lib/streams) for 68k based machines. If you were using 10.1, we encouraged upgrading the os to 10.2. The primary purpose for the 10.2 patch is to fix the pty problem. Most problems with tcp result from misconfiguration, especially when a machine has two network interfaces. You must use the explicit network interface in the rc.local file, make sure that /etc/hostname and bldt return the same name for the node. Routing tables are setup for both interfaces when they are initialized and routed should run -f (flush route tables on initialization). Another area for misconfiguration involves the netmask. Netmasks must be uniformly applied to all hosts on the wire. For example, if you have a class "B" address, the first two octets are network bytes followed by host address in the last two octets. If you do not subnet for a class "B" address, a default mask value of 255.255.0.0 (ffff0000) is applied. If you subnet the third octet, you must specify a netmask of 255.255.255.0 (ffffff00) on the explicit interface as well as all hosts on that "wire". The subnet mask "tells" tcp to consider the third octet as a continuation of the network address. The gateway will route accordingly. Hope this helps. Bob "The Renegade" Spear spear_r@apollo.hp.com "Anything worth knowing, cannot be understood" Woody Alan
rees@pisa.ifs.umich.edu (Jim Rees) (01/17/91)
In article <4f3e90d0.20b6d@apollo.HP.COM>, spear_r@apollo.HP.COM (Robert Spear) writes:
Jim, the sr10.2 version of routed is not "broken" as in 10.1. 10.1 routed was
prematurely aging routes (metric 16) and therefore, you would loose the route to
the destination. 10.2's routed would cause an access violation if invoked as
/etc/routed -t (with no filename). It worked fine in every other way.
That's not my experience, and it contradicts the patch release notes. I
don't remember what the sr10.1 routed did, but the sr10.2 routed certainly
used to forget at least some routes.
1.36 Patch m0108 /etc/routed
Patch m0108 includes fixes to the /etc/routed command for nodes run-
ning the SR10.2 version of Domain/OS. This patch is incompatible with
all other versions of Domain/OS.
Patch m0108 fixes the following problem (DDC72):
The /etc/routed command was timing out active interfaces. routed has
been modified to prevent it from timing out and thus marking "down"
interfaces that are configured "up". It does, however, time out inter-
faces which have been configured "down" via /etc/ifconfig.
Install patch m0108 on nodes running the SR10.2 version of Domain/OS
(use the bldt command to determine the revision of the operating sys-
tem running on your workstation).
Patch m0108 includes the following file:
/etc/routed 1989/11/10 20:56:21 EST (Fri)csfst1@unix.cis.pitt.edu (Charles S. Fuller) (01/17/91)
In article <4f3e1cec.1bc5b@pisa.ifs.umich.edu>, rees@pisa.ifs.umich.edu (Jim Rees) writes: > In article <79764@unix.cis.pitt.edu>, csfst1@unix.cis.pitt.edu (Charles S. Fuller) writes: > > The SR10.1 version of routed was afflicted with a form of Alzheimer's; > > The sr10.2 routed has this same problem. Either get sr10.3 or the routed > patch, number 108. sr10.1 tcp had other problems as well, you should at > least get sr10.2. > Thanks for the correction, Jim. The more I think about it, the more it seems like the problem with frequent table-flushing first appeared in 10.2. > ... The first thing I would do in this case is install sr10.2 or sr10.3. This is good advice as well. Given a choice, however, I'd go for 10.3. We've run 10.3 here for a few weeks now, and it's performed well. - Chuck