[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-users] XEN HVMs on LVM over iSCSI - test results, (crashes) and questions

Hi all,

some results from my configuration.

I've this configuration (I'm not interested in raw top performances but in reliability.. so, I can accept slow MB/sec and prefer to rely on a RAID6, for example):

1 Infortrend iSCSI Array A16E-G2130-4 with:
- 1GB DDR cache
- 7 x 500GB sataII Seagate ST3500630NS with 16mb (no budget for SAS)
- one of the logic volumes (about 1TB) is actually bounded to a single channel

1 DELL PE1950 server with:
- QLogic iSCSI HBA QLA 4060C
- 2 x Quad core x5335  2GHz Xeon
- 2 x 73GB SAS drives (raid1 software)

1 DELL 2716 gigabit switch in the middle

The array, tha HBA and the switch are Jumbo frame enabled. CHAP authentication enabled.

I've successfully installed XEN 3.1.0 from source (kernel 2.6.18-xen) and the QLogic ISP4XXX iSCSI Host Bus Adapter driver

I have successfully installed a windows XP HVM, a Scientific Linux CERN - SLC - 4 and 3 (RedHat EL 4 and 3) HVMs and other non HVM machines. The volume I imported from the iSCSI is used with LVM: each HVM domain has a 8GB drive partitioned with a 1GB swap and rest on /.

These are performance for bonnie++ (Per Chr column write and read):

iSCSI      dom0: w: 54M, r: 47M  (no domU)

Local disk dom0: w: 59M, r: 51M  (3 idle domU)

HVM single     : w: 18M, r: 37M

I've then launched bonnie++ on two separate SLC4 HVM (cloned):

HVM 1          : w:  8M,   r: 33M
HVM 2          : w:  8.5M, r: 38M

The same on three HVM:

HVM 1          : w:  4.4M, r: 11M
HVM 2, crashed, lost ssh, error: "hda: lost interrupt", need "xm reboot"
HVM 3, crashed, lost ssh, error: "hda: lost interrupt", need "xm reboot"

So, where is the limit of my configuration?

How can this scale up in your opinions in real applications (not just stressing it with bonnie++)?

I'm quite confident that HVM performance should be slow. My problem eventually is that iSCSI performance are not so bad compared on Local disk performance (which could lead to a "really poor local disk perf!").

Any suggestion on this is appreciated.

Thank you in advance,


voice: +39 010 353 2789
fax:   +39 010 353 2948

Xen-users mailing list



Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.