[openib-general] High Availability status in OFED (was Re: [openfabrics-ewg] Mellanox/Voltaire/QLogic/IBM SQA results for OFED 1.1?)
ishai at dev.mellanox.co.il
ishai at dev.mellanox.co.il
Mon Sep 25 12:31:13 PDT 2006
Hi Scott,
The IPoIB HA (High Availability)solution in OFED 1.1 is a short term
solution. (There is an on going work on a full solution, that uses
bonding).
This short term solution for IPoIB HA uses the command "ip monitor link"
to find out when a link goes down, and then updates the ip address of the
other port.
Apparently RHEL4 uses an old version of iproute package (iproute-2.6.9-3
with ip utility, iproute2-ss040831 in RHEL4.0 U4) in which there is no
unique indication when a port goes down. (It gives the same indication
when a port goes up or down).
In SLES10 there is a newer version of iproute and our solution works well
with this version.
In order to solve the problem, The next RC will include also an
installation of a version of iproute (iproute2-2.6.16-060323 with ip
utility, iproute2-ss060323). This version will be installed only for OFED
installation that includes the IPoIB HA option and only on RHEL4. The
package will be installed in a private directory inside the OFED directory
(It will not replace the iproute version of the distribution) and will be
accessed by the IPoIB scripts using the exact path.
As for SRP HA:
SRP HA is currently available only for SLES10. The reason is that SRP HA
uses the device-mapper multipath that needs high version of udev (>050).
RHEL4 uses udev 039.
Ishai
>> As for the HA it works on SuSE but not on RH. Ishai will
>> issue a report.
>
> This will be fixed for 1.1, right?
>
> Scott
>
> _______________________________________________
> openfabrics-ewg mailing list
> openfabrics-ewg at openib.org
> http://openib.org/mailman/listinfo/openfabrics-ewg
>
More information about the general
mailing list