Meeting Minutes: Wednesday Feb 25, 2015 Kick-Off Meeting after TSC Approval It was difficult to record all attendees as not all showed up with names listed in the GTM attendee window. We likley should make an effort moving forward to record the names of attendees on the call. Partial List: (pls help me add others) LingLi Deng (China Mobile) Francois-Frederic Ozog (6Wind) Bob Monkman (ARM) Keith Wiles (Wind River) Fahd Adibi (EZChip) Abdel Rabi (unknown) Saikrishna Kotha (Xilinx) Subhashini Venkataraman (Freescale) Rob Dimond (ARM) Ola Liljedahl (ARM)
LingLi asked for Bob Monkman to review the BreakOut Discussion results from OPNFV HackFest Bob reviewed LingLi's slides that provided an overview of the project, but before getting to the observations of the feedback, a long discussion ensued regarding the goals and approaches of the project described in the overview slides.
One of the essential questions centered around how aware the VNFs would be/should be regarding the capabilities of the underlying hardware, and what does it mean to say there can be a set of APIs that VNFs could leverage to access various accelerators. For example, looking at crypto interfaces, it was pointed out that some operations can be async, while others can be synchronous and it may be very difficult to have one set of APIs covering both cases. Beyond that, there can many other accelerators hardware can support, while other hardware may not have direct support in hardware and the question arises, do we have software implementations in these cases or does the VNF have to discover the capabilities of the underlying platform and call different APIs? This is just a fraction of the various questions and lots of ideas were shared over 30 + minutes that consumed nearly the entire call. Too many to capture accurately…
It was pointed out that this is a Requirements project and therefore we really need to step back on establish some fundamental requirements first, and then we can to drill down on individual attributes.
For example, what level of awareness/discovery should be provided for regarding the underlying resources? How can the orchestration and VIM layer be involved, based on workloads and SLAs, for example, in assigning resources to VMs and how do these upper layers become aware of capabilities.
Additionally, in earlier project discussions, the question of binary compatibility for VNFs, across hardware of a given Instruction Set Architecture, should we strive for (source-level with re-compile, binary?) had come up.
It was suggested that we begin by establishing some consensus around these very fundamental questions and progress through a series of defining requirements based on the answers to these questions. People were encouraged to bring agenda item topics along these lines to the next call and we can prioritize. It may take several calls to establish of the defining high level requirements that then break down into sub-requirements based upon the decision we as a team arrive at.
Apologies in advance for anything in particular I should have captured in more detail here. Corrections welcome and please come prepared to the next meeting to help identify the highest level requirements that can guide the more detailed requirements work in subsequent calls.
Bob Monkman
