> This might be, but, how do I tell? If the VM was overloaded, surely it > would present itself in the OS in some form or other?
Your users presented 'spotty' performance issues, so your peaks and small blips are the issue, this is often not recorded and is often averaged out in most monitoring. If your guests are HVM and properly equipped with drivers to reduce the overhead, then the question comes down to how many layers the instructions are flowing through to achieve processing and return to the TCP stack. I don't run anything that is "I/O" intensive in a VM container for the reason that they need direct instruction set access, not App -> Guest OS -> VM API -> HW -> (... and back). This suggests a WAIT code delaying processing, quite possibly the collection of said stats too. Your stated specs are great, well beyond what hlds/srcds needs, but I think you may be bumping into limits that are tiny, but collectively present as small spikes getting averaged out to nothingness. 64-bit I don't think will help; you can give it a shot, but I've never ran 32-bit PAE, I went to 64-bit pretty much from day 1. I'd recommend turning something off and seeing if the same behavior continues, maybe the forked L4D, or one of the TF2 servers. I run 25 forks of L4D2 on a RAID10 2xXEON socket system with 12gb RAM solid... my system never achieves 2.5 in system load... ever. This is an older hardware system, but has served me great for a long time. I, however, do not virtualize because I don't need to segment the system. ~Brian On Fri, Sep 9, 2011 at 12:57, Peter Reinhold <[email protected]> wrote: > On Fri, 9 Sep 2011 12:22:34 -0700, Brian Menges wrote: > >> I really think the fact remains that the guest dom he's running is a >> bit overloaded; and running non-optimized settings. > > This might be, but, how do I tell? If the VM was overloaded, surely it > would present itself in the OS in some form or other? > >> Namely the I/O is probably killing you (with whatever writes it may be >> achieving); Either run critical write portions in a RAM disk, optimize > > The box only runs src servers, the IO is an average of 10-15kb/second, > except when loading levels, which ofcourse "spikes" at 50-80mb. (Reported by > dstat io) > > This can hardly be described as overloaded? > > Also, the VM boxes are hosted on separate, physical harddrives, and looking > at overall ESXi IO, we're talking average disk IO across the entire machine > in the sub-mb-second range. > > > /Peter > > _______________________________________________ > To unsubscribe, edit your list preferences, or view the list archives, > please visit: > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
