This post is more than 5 years old
1 Message
0
17538
October 19th, 2011 13:00
Multipathing problem on Ubuntu Lucid
We have a new dual controller Compellent SAN. Each controller has 4 iSCSI ports. Two ports in each controller are part of a separate fault domain with a VIP. Each fault domain is serviced by a single iSCSI switch (two total). The switches are not connected to each other.
We're running the following flavor of Ubuntu: 2.6.32-33-server #72-Ubuntu SMP Fri Jul 29 21:21:55 UTC 2011 x86_64 GNU/Linux
I have installed multipath-tools and open-iscsi. I can successfully connect to the volumes that Compellent is presenting to and get the I/O performance that I would expect. I can run iftop and see that traffic is being balanced evenly between my two Intel NICs.
The problem comes during simulated failovers. If I reboot one of the iSCSI switches and bring down that fault domain, the Linux server stops sending data on the other path. It just sits there until the switch comes back up and then both physical paths come alive. That can take up to 60 seconds and that is a looooong time for I/O to be paused on a server.
It seems that I've tried every combination of settings in multipathd.conf and all of them experience this issue. An example:
defaults {
verbosity 4
user_friendly_names yes
path_grouping_policy multibus
#features "1 queue_if_no_path"
#path_checker rdac
#selector round-robin 0
failback immediate
pollint_interval 5
#hardware_handler "1 rdac"
}
blacklist {
devnode "^(ram|raw|loop|fd|md|dm-|sr|scd|st)[0-9]*"
devnode "cciss"
}
I should note that we had the same behavior on our Windows 2008 R2 server when we configured iSCSI to follow the best practices guide from Compellent. When we changed the MPIO policy from round robin to least queue depth, the alternate path stayed up the entire time. From my reading, multipath-tools only supports round robin.
How do I configure multipath-tools and open-iscsi so data continues flowing over the remaining path when I lose the other one?
Thx.
Craig
0 events found


DELL-Bill Gr
50 Posts
1
October 20th, 2011 12:00
Craig,
Copilot Support has been made aware of this issue and is working it. I will send you a PM with your case number.