De installatie van onze nieuwe cluster servers loopt tegen hele rare performance problemen aan. Ik ben benieuwd of iemand dit al eens meegemaakt heeft.
System info:
cu1:~# uname -a
Linux cu1.xxx.nl 2.6.26-2-xen-amd64 #1 SMP Thu Nov 25 06:39:26 UTC 2010 x86_64 GNU/Linux
cu1:~#
cu2:/etc/xen# uname -a
Linux cu2.xxx.nl 2.6.32-5-xen-686 #1 SMP Fri Dec 10 20:52:42 UTC 2010 i686 GNU/Linux
cu2:/etc/xen#
Fysieke informatie:
HP Proliant DL360 G^
2x Quad-core Xeon
32 GB 667Mhz Memory
2x 72GB SAS 10k rpm disks
2x 300GB SAS 10k rpm disks
2x Gbit Nic Onboard (firmware-bnx2)
cu2:/etc/xen# ifconfig
eth0 Link encap:Ethernet HWaddr 00:26:55:7b:8a:1a
inet addr:xx.xx.xxx.38 Bcast:xx.xxx.xxx.127 Mask:255.255.255.128
inet6 addr: fe80::226:55ff:xxxxxx/64 Scope:Link
UP BROADCAST RUNNING MULTICAST MTU:1500 Metric:1
RX packets:52352 errors:0 dropped:0 overruns:0 frame:0
TX packets:7377 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:2873118 (2.7 MiB) TX bytes:888460 (867.6 KiB)
lo Link encap:Local Loopback
inet addr:127.0.0.1 Mask:255.0.0.0
inet6 addr: ::1/128 Scope:Host
UP LOOPBACK RUNNING MTU:16436 Metric:1
RX packets:29 errors:0 dropped:0 overruns:0 frame:0
TX packets:29 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:2912 (2.8 KiB) TX bytes:2912 (2.8 KiB)
peth0 Link encap:Ethernet HWaddr 00:26:55:7b:8a:1a
inet6 addr: fe80::226:55ff:fe7b:8a1a/64 Scope:Link
UP BROADCAST RUNNING PROMISC MULTICAST MTU:1500 Metric:1
RX packets:285379 errors:0 dropped:0 overruns:0 frame:0
TX packets:119443 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:1000
RX bytes:328974698 (313.7 MiB) TX bytes:11139062 (10.6 MiB)
Interrupt:18 Memory:f8000000-f8012800
vif13.0 Link encap:Ethernet HWaddr fe:ff:ff:ff:ff:ff
inet6 addr: fe80::fcff:ffff:feff:ffff/64 Scope:Link
UP BROADCAST RUNNING PROMISC MULTICAST MTU:1500 Metric:1
RX packets:112293 errors:0 dropped:0 overruns:0 frame:0
TX packets:249838 errors:0 dropped:82 overruns:0 carrier:0
collisions:0 txqueuelen:32
RX bytes:8213898 (7.8 MiB) TX bytes:324917145 (309.8 MiB)
Chain FORWARD (policy ACCEPT)
target prot opt source destination
ACCEPT all -- xxxxxx anywhere PHYSDEV match --physdev-in vif6.0
ACCEPT udp -- anywhere anywhere PHYSDEV match --physdev-in vif6.0 udp spt:bootpc dpt:bootps
ACCEPT all -- xxxxxxx anywhere PHYSDEV match --physdev-in vif7.0
ACCEPT udp -- anywhere anywhere PHYSDEV match --physdev-in vif7.0 udp spt:bootpc dpt:bootps
ACCEPT all -- xxxxxx anywhere PHYSDEV match --physdev-in vif12.0
ACCEPT udp -- anywhere anywhere PHYSDEV match --physdev-in vif12.0 udp spt:bootpc dpt:bootps
cu1:/proc/sys/net/ipv4/conf/all# cat proxy_arp
1
cu1:/proc/sys/net/ipv4/conf/all#
cu1:/proc/sys/net/ipv4/conf/all# cat forwarding
1
cu1:/proc/sys/net/ipv4/conf/all#
Host file:
#
# Kernel + memory size
#
kernel = '/boot/vmlinuz-2.6.26-2-xen-amd64'
ramdisk = '/boot/initrd.img-2.6.26-2-xen-amd64'
memory = '512'
extra = "console=tty xencons=tty"
#
# Disk device(s).
#
root = '/dev/sda2 ro'
disk = [
'phy:/dev/vg/ns1.xxxxx.eu-swap,sda1,w',
'phy:/dev/vg/ns1.xxxxx.eu-disk,sda2,w',
]
#
# Hostname
#
name = 'ns1. xxxxx.eu'
#
# Networking
#
vif = [ 'ip=XX.XX.XX.XX.9,mac=00:16:3E:F5:16:21' ]
#
# Behaviour
#
on_poweroff = 'destroy'
on_reboot = 'restart'
on_crash = 'restart'
------------------------------
Alles lijkt mij goed te staan maar het valt mij op dat er enorm veel verkeer over de vif's gaat en er ook veel geblokt word. Een ping naar deze servers (DomU's) geeft een loss van 40 tot 60%!
--- XX.XX.XX.9 ping statistics ---
19 packets transmitted, 12 received, 36% packet loss, time 18021ms
rtt min/avg/max/mdev = 0.032/0.181/1.794/0.486 ms
Ik heb inmiddels geen flauw benul meer waar ik nog zou moeten kijken. Zoals je kunt zien heb ik de cu2 al helemaal geupdate/upgrade naar Squeeze terwijl de andere nog op Lenny draait. Ook dat was niet de oplossing.
Mocht iemand iets weten, heel graag !