Blog

Server cluster and availability indexes


Tags :


OK, this used to make sense and actually have some value.  But right now it seems to be working against me in 6.5.3 at a customer site.  Now we have to include the expansion factor into the equation.  These two together only measure the response time, the client could see more.  So how is it beneficial?  So they are saying the indexes help decide which server for load, but the client could see worse.  Well I was seeing both of the indexes go up and down like stocks before and after a merger.

You can adjust the indexes and expansion factor as you wish, but they were not high enough to begin with.  So we started that morning with an unstable 6.5.1 server cluster, with availability somewhere in the 40/14 range with the 40 being the internal server.  After the upgrade to 6.5.3 we saw the availability jump to 75/30 on normal load.  This tells me there was some improvements along the way in stability and scaling.  Yes, they do use iNotes very sparingly on the outside server.  Most load is the clients accessing mail and applications.


What is the point to all this?  Well the expansion factor stayed low on the outside server, around 7-8 but jumped to as high as 60 when it was the only server and we were upgrading the internal one.

My guess is that the outside server was sending traffic randomly between the internal NIC and external NIC to talk to the same server.  But wait, you say!  Chris, you said one was internal and one was external. Due to their architecture, you can get to the outside server from inside to let them have some sort of cluster.  But, since the connection records use DNS, it reads the external IP address and tries to go out through the proxy and Internet to connect to the server.  The organization does not run internal DNS and relies on the ISP.

I verified they did not have any ini parameters to adjust the availability and help regulate load.  They did not and were allowing Domino to decide the factor on the fly for them at each polling. To make this shorter, we decided to let it sit this weekend and get a better range of availability with a couple days of usage instead of relying on the few hours after the upgrades.

More on Monday or Tuesday on this topic then.