[ofw] WWG - WinOF 2.0 release meeting (7/22/08)

[Stan C. Smith]

Attendees:
  Voltaire, Qlogic, Microsoft & Intel

Summary:  WinOF 2.0 functionality freeze delayed until Aug 5th. Not cooked; see below.


Microsoft: (Fab Tillier)
------------------------

Large system scaling problems identified by MS MPI testing. Culprits are openSM/SA due to failures in processing path
record lookups in a timely manner. Similar problems experienced in Linux openSM, although mitigated somewhat by newer
version of openSM. Windows openSM vintage is somewhere close to OFED 1.2+ and is lacking major large scale performance
fixes found in later OFED openSM releases.

Fab is considering using IPoIB's ARP capabilities to create MS MPI specific path records in order to not interact with
openSM/SA. Work is slated to be completed/tested by 8-5-08.

ND will be fully supported in WinOF 2.0. The ND user mode provider (.dll binary) will be provided by a 3rd party (Fab
suggested Mellanox?) to Stan for inclusion into WinOF 2.0 installers.
ND source has yet to make it thru Microsoft legal and into the svn tree, hence the binary only distribution. As of
today, the source is still under proprietary MS contract and 3rd party with the 3rd party having rights to redistribute
binaries generated from the source.
Can the OFA/Windows redistribute this binary?


Voltaire: (Yiftah Shahar & Anatoly Greenblatt)
----------------------------------------------

Several submitted patches pending final review.

WHQL testing, some pass (i/o, NDIS) some fail (Plug-n-play & .inf files). It was pointed out mthca was WHQL'ed (same
code as in WinOF 1.0), there were WHQL exceptions granted.

Voltaire customers reported WinOF 1.1 had problems with DHCP and IPoIB multicast. Stan requested these and 'all'
customer reported bugs be entered as Bugzilla bugs so they will not be lost.

Anatoly asked what is the process to address customer identified bugs in between WinOF releases?
It's unfair and poor business practice to make customers wait for next WinOF release, how are interim releases handled?
Segway into driving signing. OFA is still attempting to purchase SW publisher certificate and cross certificate from
Authorized certification authority (VeriSign). Unbelievably slow progress!
Entire topic of bug fixes between WinOF releases tabled until next WWG meeting (8-5-08) in order for all to discuss with
peers and come to a conclusion to be presented at next WWG meeting.

Anatoly asked about documentation updates. General consensus is maintainers are responsible for user level
documentation. If maintainers are not comfortable with editing doc\manual.htm, send Stan text and he will input.

Along the same lines, maintainers need to send Stan release notes documentation text; ASAP!!!!


Qlogic: (Eleanor Witiak)
------------------------

SRP testing is looking very good, close to 100% success on all tests. Eleanor will ping Stanley Sufficool to make sure
his Bugzilla reported issues have been addressed.

Mysterious VNIC BSOD issue testing has gone 14 days without crashing; remains elusive.

Stan will supply Eleanor a svr08 installer (svn.1433) for testing; next week at the earliest.


Mellanox: not present, status reported by Stan.
-------------------------------------------------------

We have some delay in MSI-X implementation and I hope we can make it
until 5-Aug but I'm not sure of it.

IPoIB - We implemented the checksum offload for ConnectX and we are
working on implementing LSO, and checksum offload for InfiniHost III. We
have a lot of ideas on ways to further improve the performance. I guess
we will try to implement as many features as possible until the code
freeze.

I don't think we should delay the release because of these features
after the 5-Aug. If we will not finish them we will implement them to
the next release.


Intel: (Stan Smith)
-------------------

DAPL/DAPL2 private data on EP accept() bug resolved & committed.
Behind in socket-CM testing for compatibility with Linux DAPL2 socket-CM.

IBBUS filter driver work proceeding:

Single HCA support working well, resolved shutdown problem so now the entire IB stack can be right-click uninstalled and
Found-new-hardware wizard installed without reboot.
Multi HCA support under debug, IBAL PNP events need to be dispatched in the ibbus ibal pnp handler via CA guid.

Wix installers are back, thanks to Fab for the WXP32/WSD build failure fixes.
Encounter WinOF 2.0 install problems on HPC server, similar to what Mellanox had reported; under investigation.