> -----Original Message----- > From: Testbed Support for GridPP member institutes [mailto:TB- > [log in to unmask]] On Behalf Of Stephen Burke > > > in the past sites without VMs have tended to complain about needing > separate nodes for every service > I think that was mostly a problem of wasted hardware; once you start looking for proper rack-mount servers you're in for about a grand and a half just to get running, which is overkill for a tiny thing like the BDII. You don't have that problem with VMs. > so I would assume that we still want to support co-located > services where possible - or is it now reasonable to expect > everyone to use VMs? > Frankly I think it is - it's a standard well established feature of the OS now, so I don't think it's asking too much. However, what I said in the context of this thread was that "I wouldn't share nodes between services at all in a virtualized setup"; if you're going to have a VM setup anyway you might as well use it to the greatest possible effect and use it for everything[1]. If someone were to go for a VM-free site, then it would probably be silly to dedicate individual hardware to things like the BDII and cluster publisher, but I think it's now pretty silly to go for a VM-free site at all. Ewan [1] Everything that suits VMs, so maybe not things like DPM head nodes, squid servers etc. Though if the disk system is fast enough, maybe even those too.