[CentOS-virt] A few questions

Tue Jul 14 20:23:45 UTC 2009
Ben Montanelli <montanelli at rivint.com>

To begin: No "problem" per se at the moment. And thank you all for the 
tips and information on this list.

If no one objects I would like to promote myself from "newb with scars" 
to "clueless duffer."

I need to move forward on my builds and I have some xm dmesg items 
unresolved or misunderstoo.

I have 2 domU's loaded, both CentOS 5.3, one as image, one as LVM 
blocks. SELinux is On and Enforced in domUs (not yet in dom0). 
Apparently all is currently stable.

Xen version 3.1.2-128.1.16.el5
domO = CentOS 5.3 current
Hardware = TYAN S2912G2NR mobo, AMD 2 x 2212, WD VelicoRaptors in Raid1

1) Are the below messages benign? I have read all over the place and 
they are quoted in immediate lockup or failure to boot situations. I 
have had not had a lockup or crash (after cleaning up the domU confs and 
services, zeroing hwclocks, etc.) in days.

(XEN) ioapic_guest_write: apic=0, pin=2, old_irq=-1, new_irq=0
(XEN) ioapic_guest_write: old_entry=00010000, new_entry=000009f0
(XEN) ioapic_guest_write: Installing bogus unmasked IO-APIC entry!
(XEN) mm.c:649:d1 Error getting mfn 25654 (pfn 5555555555555555) from L1 
entry 0000000025654025 for dom1
(XEN) mm.c:3327:d1 ptwr_emulate: fixing up invalid PAE PTE 0000000025654025
(XEN) mm.c:625:d2 Non-privileged (2) attempt to map I/O space 000cba6b
(XEN) mm.c:3327:d2 ptwr_emulate: fixing up invalid PAE PTE 00000000cba6b025

2) I went back to HPET because there seems no harm/difference. Does this 
XM DMESG indicate Xen is "fixing" it, or do I have a loose end:

with HPET On (current state):

(XEN)  -> Using new ACK method
(XEN) Platform timer overflows in 8589 jiffies.
(XEN) Platform timer is 25.000MHz HPET

with HPET Off:

(XEN)  -> Using new ACK method
(XEN) ..MP-BIOS bug: 8254 timer not connected to IO-APIC
(XEN)  works.
(XEN) Platform timer overflows in 2 jiffies.
(XEN) Platform timer is 1.193MHz PIT

Is HPET On and not setting clock to Jiffies going to bite me at some point?

3) I have been using the Apache ab "stress test" up to the edge of where 
I get cut off with synflood protection. My Xen servers will be primarily 
web applications and WAN services. When I first started test with ab I 
was able to provoke "Time goes backwards" reactions. I tweaked Xen and 
now I cannot crash or panic anything.

Is stress testing with ab an appropriate test for my environment and I 
can feel a bit more confident and move forward on my build?