[Xen-users] Gathering "metal" stats

[Tyler Szabo <tyler.szabo@xxxxxxxxx>]

Hey All,

I’m attempting to gather stats on usage of the “metal”, by which I mean the physical host’s hardware. I would like to know the CPU, IO, and network stats for the hardware.

Normally I would get these from /proc/stat, /proc/diskstat, and ifconfig respectively. However, because of the segregation of VMs and Dom0’s special relationship as a paravirtualized I don’t know if I can trust this data.

I’m aware that /proc/stat is unusable as Dom0 will not see cycles that DomUs are stealing, and thus it does not give me accurate utilization of the metal – I know that I can calculate the actual CPU usage of each VM using “vm list –long” and comparing cpu_time to uptime, so that’s not a problem.

The issue arises when I attempt to get IO stats – DomUs see their own ‘sda1’ (or whichever) – I don’t know if this means that the ‘sda’ that Dom0 sees is the real hardware or an abstraction. The question I have is: do IO operations on DomUs result in IO operations on Dom0’s ‘sda’ such that /proc/diskstat sees it? Or is that hidden from Dom0 like DomU CPU usage?

Similarly with network stats: I’m sure that Dom0’s eth0 does not report network traffic that DomUs used, but does peth0 report traffic that ‘hit the wire’? Additionally, does this include traffic that occurs between Dom0 and DomU, that doesn’t escape into the network substrate?

I’m testing this myself as I write this, but I’d rather have my “research” confirmed by knowledgeable users.

Thanks all,
-Tyler Szabo 
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users