Hi all, I am using UCARP on two FreeBS highly available apache virtual hosts.&nb work very well except that occasionally one webserv inaccessable from outside it's subnet. I have narrowed an arp issue. When a UCARP IP becomes unavailable. start a constant ping to it from my machine which lives on a d ifferent subnet, all requests timeout. I log into the cisco router th view th listed for that clear the arp cache table update correctly aft the webserver is again accessible So, I have come up with a few include: 1.&nbs that outages a problably stress th about cisco devices so think that this is a bad th cache arp entries on just the currently consists of about 25 hosts all? 2. Run an ANT task to task can become part of the UCARP solution but security is a concern. 3. Find a way to make the FreeBSD nodes produce more than usual gratuitous arp packets or add a line to do only that in the UCARP scripts. may be able sometimes d Could it be a p properly? & Any input at all is greatly appreciated. Thanks, Colin
On Monday 19 September 2005 19:31, Colin Farley wrote:> 1.&nbs=p; Set the arp cache timeout of the cisco router very low so > that outages a=re minimal. I would rather not do this as it will > problably stress th=e router too much. Unfortunately I know little > about cisco devices so=I really cant figure this one out, does anyone > think that this is a bad th=ing? Can you tell a cisco device not to > cache arp entries on just the=internal interface? The subnet > currently consists of about 25 hosts =so this may not be so bad after > all?Depending on your Cisco router model you will not have any issues whatsoever lowering the timeout to really low, in the region of a few seconds. even an old 25xx device would be to handle that without problems.> > 2. Run an ANT task to =clear the cache on the cisco device, this > task can become part of the UCARP=scripts. This may be a good > solution but security is a concern.This would be very very bad, cause no matter how you do it the security concern would be severe. -- Matt Douhan www.fruitsalad.org (remember, amateurs built the Ark, professionals built the Titanic)
Hi Matt, Thanks for your reply. you think that lowering the seen that Cisco does not recommen after reading your reply and seeing as the hosts on this subnet I would think that thi confirm. Thanks again. Colin -----owner-freebsd-stable@freebsd To: freebsd-stable@freebsd.org From: Matt Douhan <matt@fru Sent by: owner-freebsd-stable@freebsd.org Date: 09/19 cc: Colin Farley <Colin.Farley@ecarecenters.com> On Monday 19 September 2005 19:31, Colin > 1.&nbs=p; Set the arp cache timeou very low so > that outages a=re it will > p know little> about cisco devices so=I really cant figure this one > think that this is a bad th=ing? &n cisco device not to > cache arp entri subnet > so bad after Depending on your Cisco router model you will whatsoever lowering the timeout to really low, in seconds. even an old 25xx device would be to han > > 2. Ru device, this > go > solution but security is a concern. This wou security concern -- Matt Douhan www.fruitsalad.org (remem Titanic) ___ ______________________ 5F__ freebsd-stable [1]http://lists.freebsd.org/mailman/li To unsubscribe, send any mail to "freebsd-stab le-unsubscribe@freebsd.org" References 1. 3D"http://lists.freebsd.org/mailman/li
On Sep 19, 2005, at 3:08 PM, Colin Farley wrote:> Thanks for your reply. The model of the Cisco router is 2811. Do > you think that lowering the timeout to 5 seconds would be ok? I > have > seen that Cisco does not recommend a timeout below 30 seconds but > after reading your reply and seeing as the re are only a couple > dozen > hosts on this subnet I would think that thi s would be fine. > Please > confirm. Thanks again.Remember that the router is going to have to re-ARP for these hosts whenever something external sends traffic to them, unless the router already has another active connection going. The thing is, ARPOP_REQUESTS use a broadcast MAC address which gets sent to all of the machines on the network, which adds processing overhead not just on the router itself but also on all of these machines. Fortunately, you can see what this overhead is quite easily in order to tune things: Run "tcpdump -nt arp" and see how often your Cisco is making requests with a 5-second ARP cache timeout. So long as your network is only getting, say, a single-digit number of ARP requests per second, this amount of overhead is not going to matter significantly. Adjust as needed. -- -Chuck
Colin Farley wrote:> Hi Matt, > > > > Thanks for your reply. =he model of the Cisco router is 2811. Do > you think that lowering the=imeout to 5 seconds would be ok? I have > seen that Cisco does not recommen= a timeout below 30 seconds but > after reading your reply and seeing as the=e are only a couple dozen > hosts on this subnet I would think that thi= would be fine. Please > confirm. Thanks again.It should be fine, but some experimentation will be in order, as the amount of traffic each node generates could be a factor as well. Set it to what you want, then keep an eye on CPU utilization on the router. The command you want is: show proc cpu -jav