LVS
lvs-users
Google
 
Web LinuxVirtualServer.org

Expected Failover Time and Configuration Limits.

To: <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
Subject: Expected Failover Time and Configuration Limits.
From: <ntadmin@xxxxxxxxxxxx>
Date: Mon, 15 Sep 2003 16:23:12 -0700
Hello, a couple of questions...
 
We are running a dual lvs-NAT setup (2x dual 733 dell systems with 1gb ram)
Each machine has 2 active nics one with our 216.177.xxx.xxx (public network side) and the other with 192.168.xxx.xxx (private network)
 
Redhat 9 install with modified kernel. (uname -r  > 2.4.20-18.7.hidd.ipvs109.cipe154smp)
ipvsadm --version > ipvsadm v1.21 2002/11/12 (compiled with popt and IPVS v1.0.9)
 
We are running heartbeat+mon for failover between directors.
Our haresources file currently houses 144 IP addresses.
We are also running ospf and zebra on the directors.
 
During our early testing and early production our failover was virtually unnoticable.  Since we have added ospf and zebra as well as the majority of those entries in our haresources file our failover time has hit somewhere around +/-10 minutes.  Can anybody tell me if that would be normal for the ammount of resources we are failing over or if it hints to a possible problem?  We have been able to cut this time a little shorter by clearing the arp cache on our routers but do not know if the problem is actually arp or not.  Is there an easy way to tell if the grat. arp is working?  Also we noticed something that seemed quite strange, on the virtual interfaces (eth0:##) after about eth0:42 it begins skiping every other interface number (eth0:42,eth0:44,eth0:46...) and then sometime later changes to the odd number interfaces and skips the evens.  So with 144 ips we end up with the last number in our eth0 series being eth0:237, and sometimes during a failover a few of the interfaces are not correctly brought down from the failing lvs node...  In addition to this we have found that a few (3 or 4) of the virtual interfaces with real-world IPs are responding to arp requests.  The only place on our network where these addresses exist is inside of the Load Balancers.  It is my understanding that virtual interfaces should not respond to arp requests.  Any insight or help you could provide would be greatly appreciated.  I can provide any information you might need to assist you in helping us out.
 
Thank You
Billy Olson
Systems Administrator
ReachONE Internet, Inc.
<Prev in Thread] Current Thread [Next in Thread>