[ofa-general] OFED October 15 2008 meeting summary

[Tziporet Koren]

OFED October 15 2008 meeting summary on OFED 1.4 status toward RC3:

Meeting minutes on the web:
http://www.openfabrics.org/txt/documentation/linux/EWG_meeting_minutes/

Meeting Summary:
============== 
1. RC3 is delayed - will be released early next week
2. GA date was moved to Nov 10
3. NFS-RDMA will be disabled by default and its quall level will be
technical preview
4. RDS development must move to a new branch since last changed broke
functionality
5. Three issues found in the interop event - bugs should be opened

Details:
======
1. OFED 1.4 status:
- Moved to kernel 2.6.27
- MPI: Open MPI 1.2.8 ready, MVAPICH2 and MVAPICH updated.
- uDAPL: compat-dapl-1.2.11 and dapl-2.0.14 released
- NFS-RDMA critical bug with NFS_RDMA on SLES 10 SP2 - still on work -
should be fixed for RC3
  Decision: NFS-RDMA will not be installed by default. Jeff will see if
this can be a separate RPM
  Reason: This is a new component that was not tested enough 
- OSM: Cashed routing - Should be checked in once patches V2 will be
ready (patches were sent yesterday)
- RC3 schedule: Going to build it on Sunday, and release on Monday
- RDS: We see many changes in RDS that break functionality - need to
open a stable branch for the release

2. Rupert reported on three issues found in the Interop event in UNH :
- IPoIB - issue with packet loss when package size larger then 8196
bytes
- SRP - issue with one vendor on large table size
- link is not us if OSM is not running for some cards
Rupert will make sure that bugs will be opened

3. GA date: Decided on a new target date: 10 Nov 2008
This means the Logo program at UNH will be done after sc08

4. Bugs review:
1128    	blo  	Othe  	stefan.roscher at de.ibm.com  	release
IPoIB-CM QP resources in flushing CQE context
1278 	blo 	RHEL 	perkinjo at cse.ohio-state.edu 	Mvapich2 fails
to compile when uDAPL option is specified
1242 	cri 	RHEL 	yannick.cote at qlogic.com 	kernel panic
while running mpi2007 against ofed1.4 -- ib_... - should be for rc3
1257 	cri 	All 	eli at mellanox.co.il 		Severe
performance penalty for PCIe strict ordering - on work
1262 	cri 	Othe 	andy.grover at oracle.com 	congestion hang with RDS

1164 	maj 	SLES 	yosefe at voltaire.com 		iperf over IPoIB
fails for 100 tcp connections - on work
1221 	maj 	SLES 	Jeffrey.C.Becker at nasa.gov 	SLES10 sp2:
remote logins via ssh fail due to rpcbind and...
1248 	maj 	SLES 	monis at voltaire.com 		Bonding - after
reboot the host stucks while raising the ... - not reproduced
1282    	cri  	RHEL  	amirv at mellanox.co.il  	  	Kernel
panic during Netperf run  


Tziporet