Re: Xen and iscsitarget

[Alvin Starr <alvin@iplink.net>]

Nathan Lutchansky wrote:

>Hi all,
>
>I've been using Xen for a few months now and I'm quite impressed.  Not
>only is the virtualization platform stable and well-designed, but the
>surrounding administrative tools and processes are usable and complete.
>(Although I admit, I was kind of disappointed with the performance until I
>realized recently that I hadn't compiled support for my ATA chipset into
>the dom0 kernel and all disk I/O was PIO.)
>
>My problem is that I'm unable to get the Cisco linux-iscsi initiator 
>running under Xen.  It seems other people have been able to do this so 
>maybe I'm doing something wrong.
>
>I'm running the Xen 2.0 release candidate (which, BTW, is very nice 
>otherwise, I've had no other problems) and have iscsitarget 0.3.4 running 
>on an external host.  The iscsitarget has been working fine with non-Xen 
>hosts.
>
>Unfortunately, I'm unable to get unprivileged domains to talk to the 
>iscsitarget.  Using Linux 2.4.27-xenU with linux-iscsi 3.6.2 I get the 
>error "xmit_data failed to send 8240 bytes, rc 48", and using Linux 
>2.6.9-xenU with linux-iscsi 4.0.1.10 it simply locks up the whole kernel 
>on session initiation.  (Pings still work though.)
>
>In domain 0, I can use linux-iscsi-4.0.1.10 to talk to the target 
>successfully, however, under load I will often get syslog messages 
>"iscsi-tx: page allocation failure. order:0, mode:0x20" and a big call 
>trace (see below).  These apparently cause no real problems though, and 
>all my data seems to be intact and the filesystem works fine.
>
>Domain:     Kernel:        Linux-iscsi:     Result:
>unpriv      2.4.27-xenU    3.6.2            "xmit_data failed"
>unpriv      2.6.9-xenU     4.0.1.10         immediate lock-up
>domain 0    2.6.9-xen0     4.0.1.10         works fine, with strange errs
>
>The lock-up with 2.6.9 happens after less than a kilobyte of data has been 
>transmitted in either direction.
>
>My unprivileged domains otherwise work fine and I've pumped many 
>gigabytes of data between them and the network.
>
>Since this only breaks from an unprivileged domain, I'm guessing that 
>there's some incompatibility between how VIFs are implemented and how 
>linux-iscsi wants to use them.
>
>Can anybody suggest kernel and linux-iscsi versions that seem to work 
>correctly?  -Nathan
>
>  
>
I am running a Fedora Core 2 as the OS on disk for Xen0
The /proc/version string is:  Linux version 2.6.8.1-xen0 
(root@xenmaster1) (gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)) 
#1 Fri Oct 15 17:47:44 EDT 2004
the version of iscsi is: iscsid version 4:0.1.10 ( 8-Oct-2004)

I am not saying that this is a fact but I have found that:
1) run iSCSI over gigabit ethernet only.
2) insure flow control is turned on for all the switches and NIC's
3) use only e10000 or sysconnect NIC's

I have had lots of weird problems with via-rhine and 100BT connecitons.

I am using the IET(iscsitarget) 0.3.2

I am importing 3 target sytemss as an array and having the XenU's do the 
actual raiding of them. So far So Good. Knock on wood. Keep my fingers 
crossed





-- 
Alvin Starr                   ||   voice: (416)585-9971
Interlink Connectivity        ||   fax:   (416)785-3668
alvin@iplink.net              ||




-------------------------------------------------------
This SF.Net email is sponsored by:
Sybase ASE Linux Express Edition - download now for FREE
LinuxWorld Reader's Choice Award Winner for best database on Linux.
http://ads.osdn.com/?ad_id=5588&alloc_id=12065&op=click