AB Meeting/June 2015 Minutes

From Xen
Jump to: navigation, search

Attendees

- Lars Kurth
- Major Haydn (Rackspace) - standing in for A Messerli who is on Vacation
- Konrad R Wilk, Daniel Kiper (Oracle) 
- Donald Christopher Slutz (Verizon / Terremark) 
- James Bulpin (Citrix)
- Larry Wikelius (Cavium) 
- Philippe Robin (ARM) 

Quorum

Note: Sarah tried to dial in and got an engaged tone. Please let me know if you have issues next time. I raised a support ticket, but it was too late (issues need to be raised within 24h)  
*If anyone else had difficulties – I know Sherry was struggling last month – please let me know*

Agenda

Operational Issues / Questions

Face 2 Face BoD Meeting

Face to face board meeting August 19th (instead of 18). Agree exact time due to overlap with LinuxCon
Who is at LinuxCon & CloudOpen: James, Konrad, Major, Thomas M, ... possibly some more

** We discussed what time is best and agreed that we will do the meeting on the 19th from 9:00 - 10:00 PCT **
Lars will update the invite - the meeting is early enough to dial in people from Europe, if needed
Breakfast will be provided

Open Actions from last two meetings

ACTION: Lars will follow up with Steve Westmoreland re MS Window licensing for Test Lab (in progress, but have no final proposal yet)

Update from Lars:
* Current MS proposal not practicable as it requires that an unclear number of developers will require MSDN subscriptions
* Asked for clarification with some specific examples to figure out who would need subscriptions
* Also asked for per VM subscription model as al

ACTION: Konrad to recommend any competent sysadmins in the greater Boston area 
* Have started to reach out to several possible candidates and put together a set of interview questions and tests
* Tim Mackey who is based in the area is also reaching out to misc candidates

ACTION: Philippe get back to Lars/the board with more information (re possible ARM HW for the next phase of Test Lab expansion)
* Lars is following up on any necessary paperwork that may be required for APM HW donation
* With AMD and Cavium on the AB, we will eventually have boards from AMD and Cavium too
* The question Lars had to Philippe is whether this will give us enough coverage 

Approvals

In our 2015 budget, we have a line item for hosting the OpenStack Xen Project CI loop


                 OpenStack CI Support = Total in 2015
2015.9  6700     Hosting cost for CI Loop = 15000.00

Rackspace has kindly volunteered to make this available for free until the CI loop is set up and running. This has now been completed. The estimated cost for running the loop are on average $1500 per month (see http://wiki.xenproject.org/wiki/AB_Meeting/May_2015_Minutes#OpenStack_Update_.28Lars.29) and I would like to get the accounts moved over to the Xen Project from beginning of Q3. The estimated annual cost is $1.5K per month (or $18K per annum). This would leave us an estimated cost of $9K vs $15K we budgeted for this year.

RESOLUTION: Approve setting up Rackspace accounts for Hosting the XenProject - LIBVIRT - OPENSTACK CI Loop using the budget line item above

**Approved**

Reminders

* Developer Summit sponsorships: http://events.linuxfoundation.org/events/xen-project-developer-summit/sponsors

Currently have ARM and Citrix sponsorships
ACTION: Lars will follow up on whether there are any other ongoing discussions

*Developer Summit registration: http://events.linuxfoundation.org/events/xen-project-developer-summit/attend/register

Escalations

PVH Progress

* A number of maintainers asked me to escalate lack of resourcing and progress on the PVH feature. At the Developer Meeting in San Diego in 2012, the rough plan we agreed to was to aim to replace PV with PVH (aka deprecate PV and have PVH in production ready by 2017). 

This now looks at risk: there has been very little progress on PVH in this release cycle.
** The relevant thread is http://lists.xenproject.org/archives/html/xen-devel/2015-06/msg00258.html
** Concerns raised were in particular about the expectations that were raised in the previous releases and also via talks at LinuxCon/CloudOpen this year

1) I would like to find out how important PVH is for AB members

James: Personally on the fence with this question. 
James: On one side: PVH as a replacement for PV is important. PV had its day. Does not fit well with modern HW. We are not taking advantage of all HW features
James: However: if we look at it from PV on HVM is good enough and much more straightforward to take advantage of
James: In summary, I am not too worried about the status quo. But I do agree that we need to decide which way we are going to go
Lars: Just to clarify, you are saying we need an achievable plan to introduce PVH or look at alternatives?
James: Correct

Konrad: From my standpoint there is also the political issue with PV - Linus and kernel maintainers want PV infrastructure removed in the long run. 
Konrad: We need to demonstrate progress
Konrad: If we don't get something done, Linus and maintainers will eventually rip out Linux support for PV

Konrad: Besides that, there is a food business cases for PVH. PV on HVM does give some benefits, PV/PVH has certain functionality that is beneficial to Oracle 
Konrad: (e.g. lesser attack surface due to no QEMU, performance boost in some areas, latency).

Major: Can say that PV on HVM is working well - but getting rid of QEMU and improving security is interesting
James: Likes getting rid of QEMU too - maybe there is a middle ground on how we can achieve this

Konrad: States that there is the impression is that little progress has been made in the last 2 release cycles
Konrad: That is not necessarily true: Oracle has some prototypes that are not yet upstreamable
Lars: Maybe post as RFC's and coordinate better
 
Konrad: Also voices frustration about feature creep
Lars: To some degree this is partly a consequence of the long time period and working against a changing baseline

Slutz: Prefer not to have the security exposure, but on Windows PVH is not a solution. Most of our workloads are Windows.

James: PVHVM is the only realistic Windows option right now
James: However PVHVM is just what KVM does - PVH would give us a long-term differentiator and will allow us to stay ahead of time
James: However, in my view as long as there is progress we do not need to change priorities

*Summary: the consensus appears to be that PVH is important, but that the current rate of slow progress is not an issue* 
*Agreed*

Konrad: Also points out that our original plan was never to deprecate PV in 2017, but to deprecate PV 5 years from when we declare PVH stable.
James: Agrees with this 

*All agreed: The 2017 timeframe is not that important as long as we can show progress.*

2) Are we happy with the current rate of progress? 

James: do not have enough insights on the question
Lars: what's your current gut feel on when it is finished, aka when we can start the deprecation period
Konrad: we are hoping to be finished with the TODO's in the code in 3 months, the AMD part in 6 months
Konrad: we will have to redo some of the Linux dependencies that's another 3 months
Konrad: then another 3-6 months for stabilisation and bug fixes
Konrad: so realistically with freeze times we are looking at 12-15 months from now; possibly we can freeze APIs at the end of this year or early next 
Konrad: In other words 12-15 months before we can declare that PVH is supported
Konrad: the 5 year clock would be ticking from this point onwards

3) Would any other vendor be willing to step up help resolve resourcing? If not, are there alternatives?

Konrad: Scope of work is very low level. It will require for someone new to get up-to-speed. 
Konrad: Using an intern or someone else, will most likely not be effective unless they have prior experience.
Konrad: Would be great if we could add more resources, but this seems to be unfeasible *unless* we are talking about someone who can get up to speed immediately
Konrad: Even if we did, we would probably not improve the elapsed time to achieve this much.
Konrad: Is thanking Citrix that Roger is helping with PVH

Lars: Are there any other issues?
Konrad: re-design of boot API throws a wrench in the time-line 
Lars: Who has been doing this
Konrad: Andy Cooper ...  

[Correction added later: as well as Jan Beulich (Suse), who I was told accepted the PVH contribution under the condition that the boot API would be fixed 
[Additional Info: Also, it seems that Jan and Juergen Gross (both Suse) may be willing to help in the next release cycle]

ACTION: Lars to get some of the key people into a room at the Developer Side to try and resolve the details ... including some of the necessary Linux folks
* Ensure that *all* stake-holders, including Roger can participate remotely
* Also make sure that there is an ongoing heart-beat (this has worked well for the OpenStack work)

Developer Training

A number of vendors have been asking me to step up training of future developers (see line item below), in particular because OutreachY (formerly OPW) seem to have lost steam since moving to Software Conservancy and the program is now much less well run than before. Also, with big projects such as Linux not being guaranteed GSoC slots any more, we may want to look at alternatives.

We do have some budget, which we could re-use

                  Developer Programs = Total 2015
2015.5    6440    Reserve for Developer Events = 30000.00
2015.6    6210    Travel Stipends and similar = 5000.00
2015.7    6800    OPW and other education programs (4 interns) = 26000.00

Konrad: the Xen Hackathon in China has opened up my eyes 
Konrad: Chinese companies want to increase the pool of Xen developers and there is a lot of momentum for Xen right now
Lars: That is correct

Konrad: Also does not know of internship programs in China except for GSoC
Lars: Note that we did have - to make this really feasible in China we need local mentors. 
Lars: I think we need probably another 6 months of confidence building of the tranche of newer developers in the comm
Lars: I believe we can build upon recent successes in China

Lars: We could also do this ourselves ... Tizen has run a similar program to GSoC / OPW. 
Lars: Ideally we would collaborate with other Collaborative Projects to avoid overheads 
Konrad: very much likes this idea 
Konrad: I will be presenting a panel at Intel with Greg KH (could present that at next stepping stone for interns)
Konrad: if we do this, there should be focus on emerging markets 

ACTION: Lars to do some groundwork and see what the possibilities are - discuss with Mike Woster and Laura Kempke
ACTION: Lars to prepare a more concrete proposal after LinuxCon, etc.

AOB

Informational

* Modification of Security Process: discuss list has been operational for a month, process change http://lists.xenproject.org/archives/html/xen-devel/2015-06/msg01202.html implemented
* Development Updates: http://wiki.xenproject.org/wiki/Xen_Project_Hypervisor_Roadmap/4.6
* Developer Summit schedule: http://events.linuxfoundation.org/events/xen-project-developer-summit/program/schedule 
* Reached out to Clear Containers team (Intel) and Hyper_ ( https://hyper.sh/why-hyper.html) which both are solutions that run Docker in Virtual Machines to make Docker more secure