Friday February 20, 2009

How many users can you get on a lab box ?

Virtualisation holds much promise for me as a lab manager. In the lab we are taking two approaches. The exciting thing is the speed at which this technology is advancing - it is honestly difficult to keep up! Here is one of our solutions.

In our environment an engineer books a server, loads his OS of choice and and additional software and tries to reproduce the customer problem he is having. The engineer may want extra hardware added like a fibre card and associated storage. In that case - for now virtualisation is not for them. Of course there is a 'delay' waiting for the OS to install because you do not know what has been done to the OS the previous user left behind. We are taking a T5240 server with 64gb of memory and using LDOMS. You define a control domain and use enough internal disks to make a ZFS pool which will be used for the LDOMS boot disks. In our case the control domain has eight VCPU's, a decent amount of memory and so on. We then define a set of identical LDOMS. Each LDOM is given two 'blank' ZFS volumes that the LDOM sees as internal boot disks. The engineer can then install/configure the OS however they wish. We give them two disks so they can do mirroring, upgrades etc. Then using ZFS snapshots/clones each LDOM has another disk with say Solaris10U6 on it which we can then 'refresh' back to a known good clean state after the engineer finishes with it. The neat thing is of course there is no install time to wait for.

All the consoles of these LDOMS are available to users and it just feels like any lab box. We don't give engineers access to the control domain as they might end up messing up their/other folks LDOMS. We do have to provide an ability to reset LDOMS as if the engineer does a 'init 5' the LDOM will be off until it can be started by a priviledged user.

The trick of course is to get engineers to use this instead of traditional machines. We can eliminate several large servers like V880 saving space, power and cooling. We will have to keep some of these servers for product support and their smaller brothers like the V240 for when engineers want cards inserted as I mentioned above. At the moment we are quite careful on the loading of LDOM's on a server we are only using half the VCPU's/memory of the box. One my next posting on Lab Virtualisaton I will talk about the second solution we are using.