[lug] Linux boxes drop off the net? Router problem?

[D. Stimits]

I haven't seen this problem myself, but you might be interested in
installing "ntop", a kind of "top" for networks. Basically set it up so
you can see things while they work, then compare it when they stop
working.

"Gary Frerking (TurboPower)" wrote:
> 
> I've got a problem that's been driving me nuts for some time now.
> 
> I'm afraid a reasonably full explanation will make this a long message, so
> I'll summarize the problem first, then go into more detail.
> 
> In a nutshell, we have a mixed network consisting primarily of Linux boxes
> and Win2k boxes. On a seemingly random basis, our Linux boxes cannot be
> "seen" from the Internet. In other words, there are periods of time when a
> given Linux box cannot be pinged and/or connected to via any of its open
> ports. Checking the affected box locally reveals that it's fine (not
> overloaded, not rebooting, responsive locally, etc). But no one can connect
> from the outside.
> 
> We do not see this phenomenon with any of the Win2k boxes (which obviously
> doesn't help my cause any).  :o(
> 
> Our SysAdmin (a MS-friendly kinda guy) simply tells me "gee -- it must be
> Linux, none of our Win2k boxes do that").
> 
> This seems to affect every Linux box on our network. Not every box is set up
> the same and not every box was set up by the same person, so I would tend to
> think it's not a simple configuration problem.
> 
> Not every box uses the same distro either (some are RedHat 7, some are SuSE
> 7).
> 
> All boxes have the applicable patches/updates applied. All kernels are
> fairly recent 2.2.x builds.
> 
> Some of the boxes have decent hardware (P450/128 Meg), so I think we can
> rule out the boxes being "underpowered".
> 
> There are no "hosts.deny" or firewall issues getting in the way.
> 
> It's not a DNS problem
> 
> There is enough mixing in our network that I think we can rule out a simple
> hardware problem (like a bad hub, or bad wiring, or bad NICs found only on
> the Linux boxes etc). In other words, our Linux boxes don't all connect to
> the same hub or have identical hardware or anything.
> 
> -----
> 
> I'm suspecting it's more of a routing problem of some sort, but I don't know
> enough about routing/routers to know exactly *what* the problem is or *how*
> to troubleshoot further (and the SysAdmin isn't really willing to put time
> into the problem unless I come up with a specific plan of attack).
> 
> Is it possible, for instance, that the Linux boxes go into a "dormant" mode
> after a while and the router thinks they're off the net or something??
> 
> Our network topology is as follows:
> 
> A T-1 feeding a Cisco 2516 router.
> 
> The router dumps into a 3Com LinkSwitch 1000.
> 
> The switch feeds 3Com 10 meg hubs.
> 
> All of the Linux boxes are normally connected to the hubs (there was a
> period of time where we connected a Linux box directly into the router for
> testing, and we still had the same problem).
> 
> Does this ring a bell with anyone? I've posted questions (and looked for
> answers) in other places with no luck.  :o(
> 
> Any pointers on how to fix (or at least how to troubleshoot) would be much
> appreciated. We've been getting customer complaints for some time about this
> -- and last week I experienced it first-hand while trying to connect to
> various boxes from LinuxWorld.
> 
> -- Gary
> _______________________________________________
> Web Page:  http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug