[ewg] EWG/OFED Mar 8, 2010 meeting minutes

Tziporet Koren tziporet at mellanox.co.il
Tue Mar 9 08:24:07 PST 2010

EWG/OFED the meeting minutes for Mar 8, 2010

Meeting summary:
- OFED 1.5.1 RC4 is planned for Thursday, Mar 11.
  GA is planned for middle of next week
- Providing OFED Binaries for end users - EWG approved this request
- Next OFED plans: 
  - Decided to have 1.5.2 in Q2
  - 1.6 - had discussion to delay to Q1 2011, but plans will be closed at Sonoma

Meeting details:

1. OFED 1.5.1

Release schedule proposal:
- RC4 - Mar 11, 2010
- GA  - Mar 16, 2010

- Decided that NFS-RDMA will not be supported on SLES10-SPX
- We wait with RC4 for NFS-RDMA and SRP fixes.
- Changes to Chelsio FW can go in after RC4, and we will wait with the GA for it.
- RHEL 5.5 is not officially supported by 1.5.1. 
  Backports were added so people can test it but we are not declaring it is a supported OS of 1.5.1

Bugs we decided to fix for the GA:
1964  blo  	swise at opengridcomputing.com  	cxgb3 fails openmpi branding - 
1959 	cri 	tom at opengridcomputing.com 	[OFED-1.5.1 -NFSoverRDMA] - NFSoverRDMA client hits kerne...
1961 	cri 	tom at opengridcomputing.com 	[OFED-1.5.1- NFSoverRDMA] - System hits kernel panic whil...
1789 	maj 	tom at opengridcomputing.com 	OFED-1.5 kernel panic observed while running iozone & con...
1919 	maj 	tom at opengridcomputing.com 	Connection Up-Down during data transfer
1965 	maj 	yevgenyp at mellanox.co.il 	Bonding mlx4_en: ping does not resume after failover betw... 
1978  maj  	vu at mellanox.com  	  	  	Kernel Panic when unloading ib_srp

2. Request for producing OFED Binaries for end users
This is a request of XWG that think there will be a usage for having binary images of OFED
Moni from Voltaire suggested to contribute their install scripts

Pros: Fast installation, no dependency in development packages, easier for the end customer
      This will enable for the OFA to have one generic binaries for all HW supported 
      (versus the vendors packages that supports only their HW)
      Will be good for the Logo program
Cons: This can add testing & QA burden on the companies

Decided that the binary images will be based only on the GA version, thus there will be no more QA efforts. 
UNH will produce the binary images.

3. OFED plans:
Have more dot releases throughout this year and delay the new kernel base for only later

Cons: We have many patches already in the fixes directory, and moving to the new kernel 
      will eliminate the need for them.
Pros: Gain the stability of OFED 1.5.x
      Changing kernel base is reflected to all backports, and it can take few months till 
      we have a stable version again. 

Plan for 1.5.2 on end of Q2 (date to be closed in Sonoma)
- Add new OSes (RHEL 5.5, SLES11 SP1)
- Critical bug fixes
- Update the management package
- Important also for the logo program, so vendors will have a chance to fix issues found in the interop event
- It is possible to add also add-on package that does not harm others, or new low level drivers for new HW.

Everybody agreed we will need 1.5.2 release.

Suggested features for OFED 1.6:
- kernel base 2.6.35 or 36
- SRIOV support (for KVM virtualization)
- Mellanox Vnic for BridgeX
- New HW from vendors (if any)
- New scalability features (if any)
- Soft-RoCEE
- 3d torus support by OpenSM
- New management features
- mmu notifier for MPI - something that will be accepted by the kernel
- Improve connection management scheme?
- More features - will be discussed in Sonoma

Schedule for 1.6 will be closed in Sonoma (CU all there :-)


More information about the ewg mailing list