XEN 3, memory and CPUs

[Curt Sampson <cjs%netbsd.org@localhost>]

On 2008-01-23 11:48 +0100 (Wed), Manuel Bouyer wrote:

> That's expected, it's the pages that are transfered between domains for I/O.

That's interesting; I hadn't realized that Xen does this.

> 50MB may be a bit short, especially if you have lots of domUs. I would
> expect 100MB to be enough, but I've always had more than 200MB free on
> my Xen servers.

So I should not be allocating almost all the memory in the system to the
domains, but should leave 50-200 MB unallocated (as shown by xm top just
after boot) for this this I/O usage?

Are there any general rules for this sizing, based on either number of
domains and expected amounts of I/O, or observation of a running system?
What is likely to happen if I don't leave enough memory? Just slower I/O?

As well, since I'm on the topic, I understand that NetBSD 4 doesn't
have any support for multiprocessing under xen. My domUs have one CPU
allocated according to xm top, and the dmesg shows a single vcpu, as
expected. However, my dom0 has two CPUs allocated, according to xm top,
and the dmesg, though it shows one vcpu, also shows before that:

    cpu0 at mainbus0: apid 0 (boot processor)
    cpu1 at mainbus0: apid 1 (application processor)

What's the meaning of this?

> > I recompiled the dom0 kernel to having 64 vnd devices and the default of
> > xbd devices. In my /dev I generated 16 of both.

This is not really related to xen, but I was wondering about this the
other day: the standard kernel config files just say "pseudo-device
vnd", without a number. I'd thought that meant they were dynamically
allocated, as with devices such as 'tun', but a vnconfig shows four
devices; is this a different case, where the default is four if a number
isn't specified?

cjs
-- 
Curt Sampson       <cjs%starling-software.com@localhost>        +81 90 7737 
2974   
Mobile sites and software consulting: http://www.starling-software.com