Re: [Freeipmi-users] ipmipower

[Al Chu]

Hey Kevin,

That's interesting.  It seems as though the BMC is actually trying to
resolve IP addresses on the BMC before responding to a packet.  Most
BMCs that I've seen seem to blindly respond to whatever IP/MAC sent the
packet, which in my opinion is the right thing to do (keep the BMC
stupid).

> > Doing a tcpdump while doing an ipmiping shows the bmc, at roughly
every
> > > two minutes, makes an arp request towards the machine running ipmiping
> > > and drops all packets on the floor while it is waiting for the arp
> > > response. So every bmc drops some packets. Also, when starting up
> > > ipmiping, it drops some packets while it does an initial arp request.

I just realized something.  Aren't the arp responses coming back
quickly?  If the arp response is coming back in a reasonble length of
time (lets say < 1 second), the ipmi software should detect dropped
packets and resend appropriately.

I'm a bit stumped.  Does IBM have some provided software that works?  I
suppose it's possible I'm doing some subtle differences in the protocol
(maybe related to sequence numbers) that confuses the BMC while other
IPMI software doesn't.

Al

On Tue, 2007-12-11 at 09:33 -0800, Kevin Fox wrote:
> On Mon, 2007-12-10 at 18:01 -0800, Al Chu wrote:
> > Hey Kevin,
> 
> Hi Al,
> 
> > I've never played with the IBM BMCs before, so I can't be 100% sure.  
> > 
> > It does seem as though the core issue is an arp issue.  Ipmipower is
> > likely losing packets and timing out in sessions.
> 
> Exactly.
> 
> > When you look at your arp table (/sbin/arp -a), are the entries correct?
> 
> Yup. Stuff like:
> sc-bmc76 (192.168.1.76) at 00:1A:64:11:24:73 [ether] PERM on eth0
> 
> 
> > Are they possibly changing?  At the beginning and after the arp is sent from
> > the remote BMC?
> 
> Doesn't appear to be changing. Since I fixed the mac addresses
> in /etc/ethers and arp -f, it should stop talking to me all together if
> it changed. After the bmc gets the arp response back from the calling
> host, it starts talking again on the same mac address. Which leads me to
> the, "bmc dropping all packets while it does an arp request" theory.
>
> > I could see a situation where your BMCs are configured with the wrong IP
> > and/or MAC address, and thus advertising IP -> MAC address mappings
> > incorrectly. So it gets cached incorrectly for some period of time
> > (leading to packets sent to the wrong location and subsequent time
> > outs), but is later corrected by some other mechanism on your network
> > (normal IP traffic).  This is just a guess at this point.
> 
> Shouldn't be an issue since the IP->MAC mapping is permanent in the arp
> table.
> 
> > Perhaps it'd be a good idea to double check your BMC configuration to
> > make sure the IP/MAC/etc. are configured correctly.  You can use bmc-config
> > --checkout to check on this.
> 
> Looks good to me.
> 
> > Also, are you using a different IP for the BMC than what you do for the
> > normal communication?  Perhaps something could be confused in there too.
> 
> Yup. The ip's dedicated to the BMC. The compute interface has a
> different subnet/hardware.
> 
> Thanks for the help,
> Kevin
> 
> > 
> > Hopefully that's a starting point.  PLMK what you find out.
> > 
> > Al
> > 
> > On Mon, 2007-12-10 at 17:44 -0800, Kevin Fox wrote:
> > > Having problems with ipmipower. I have a 192 node cluster and am trying
> > > to use it with Powerman/ipmipower. Powerman -q is showing a large,
> > > random assortment of of nodes in the unknown state. It changes each run.
> > > 
> > > I went to the underlying ipmipower commands and stat in interactive mode
> > > comes up with a random set of timed out nodes.
> > > 
> > > The man page refers to setting fixed mac addrs in the arp tables, and I
> > > tried that.
> > > 
> > > Doing a tcpdump while doing an ipmiping shows the bmc, at roughly every
> > > two minutes, makes an arp request towards the machine running ipmiping
> > > and drops all packets on the floor while it is waiting for the arp
> > > response. So every bmc drops some packets. Also, when starting up
> > > ipmiping, it drops some packets while it does an initial arp request.
> > > 
> > > I think this is what is confusing ipmipower. I've tried lots of
> > > different settings for ipmipower but have been unable to find a set of
> > > options that come up with a reliable stat. Any ideas what I should set
> > > things to? (These nodes are IBMx3550's using an RSAII if that helps)
> > > 
> > > Thanks,
> > > Kevin
> > > 
> > > 
> > > 
> > > 
> > > _______________________________________________
> > > Freeipmi-users mailing list
> > > address@hidden
> > > http://lists.gnu.org/mailman/listinfo/freeipmi-users
-- 
Albert Chu
address@hidden
925-422-5311
Computer Scientist
High Performance Systems Division
Lawrence Livermore National Laboratory