[openstack-dev] [OpenStack GSoC] Chenchong Ask for Mentoring on "Implement a cross-services scheduler" Project

Yathiraj Udupi (yudupi) yudupi at cisco.com
Wed Mar 5 18:53:09 UTC 2014

Hi Chenchong, Fang,

I am glad that you have expressed interested in this project for GSoC.  It is a big project I agree in terms of its scope. But it is good to start with smaller goals.
It will be interesting to see what incremental  things can be added to the current Nova scheduler to achieve cross-services scheduling.
Solver Scheduler (https://blueprints.launchpad.net/nova/+spec/solver-scheduler)  has been pushed to Juno, and that BP has a goal of providing a generic framework for expressing scheduling problem as a constraint optimization problem, and hence can take different forms of constraints and cost metrics including the cross-services aspects.
So it is good to not limit your ideas with respect to Solver Scheduler BP, but in general also think of what additional stuff can be added to the current Filter Scheduler as well.

For GSoC, I don’t think you should worry about the feature freeze for now.  You can propose ideas in this theme for GSoC, and we can eventually get it upstream to be merged with Nova/ Gantt.

The cross service scheduling BP for Filter Scheduler enhancements  is here - https://blueprints.launchpad.net/nova/+spec/cross-service-filter-scheduling  We can probably use this for additional filter scheduler enhancements.


On 3/5/14, 10:33 AM, "Chenchong Qin" <qinchenchong at gmail.com<mailto:qinchenchong at gmail.com>> wrote:


Sorry for not cc openstack-dev at first (haven't got familiar with OpenStack's GSoC
custom... but it's quite a different flavor compared with my last mentoring org). I just
sent it to the possible mentors. But it turns out that openstack-dev gives lots of
benefit. :)

I noticed that Fang also has interests towards this idea. It's strengthened my thought
that it's a great idea/project.

Russell and dims showed their concerns that the project described it is far too large
to be able to implement in one GSoC term. In fact, I hold the same concern, so I
asked the possible mentors about it at the end of my last mail.

This project appear to have a "big name". But when we dig into detail of the project
description, it seems that the project is about implementing a nova scheduler that
can take information from storage and network components into consideration and
can make decisions based on global information. Besides, Sylvain also mentioned
that it's now in FeatureFreeze period. So, I think maybe we can move this project
from Gantt section to Nova section (with the consent of original project proposers),
and further specify the contents of the project to make it a enhancement or a new
feature/option to nova's current scheduler.

Thanks all your help and Sylvain's reminder on #openstack-meeting!



---------- Forwarded message ----------
From: Chenchong Qin <qinchenchong at gmail.com<mailto:qinchenchong at gmail.com>>
Date: Wed, Mar 5, 2014 at 10:28 PM
Subject: [OpenStack GSoC] Chenchong Ask for Mentoring on "Implement a cross-services scheduler" Project
To: yudupi at cisco.com<mailto:yudupi at cisco.com>, dedutta at cisco.com<mailto:dedutta at cisco.com>

Hi, Yathi and Debo

I'm a master student from China who got a great interest in the "Implement a cross-services scheduler"
project you put in the Gantt section of OpenStack's GSoC 2014 idea list. I'm taking the liberty of asking
you as my mentor for applying this project.

My name is Chenchong Qin. I'm now in my second year as a master student of Computer Science at
University of Chinese Academy of Sciences. My research interests mainly focus on Computer Network
and Cloud Computing. I participated in GSoC 2013 to develop a rate control API that is 802.11n features
aware for FreeBSD (project homepage<https://wiki.freebsd.org/SummerOfCode2013/80211RateControl80211nExtensions>). I've been following closely with OpenStack since last year and
have done some work related to network policy migration. I'm familiar with C/C++ and Python, and have
also write some little tools and simulation programs with python.

When I first saw your idea of implementing a cross-services scheduler, I determined that it's a necessary
and meaningful proposal. I participated in a research project on channel scheduling in a distributed MIMO
system last year. From that project, I learned that without global information, any scheduling mechanisms
seemed feeble. I‘ve read the blueprints you wrote and I highly agree with you that the scheduler should be
able to leverage global information from multiple components like Nova, Cinder, and Neutron to make the
placement decisions. I'm willing to help with the SolverScheduler blueprint both during this GSoC project
and after.

And, I also got a question here. According to the project description, "This project will help to build a
cross-services scheduler that can interact with storage and network services to make decisions". So, our
cross-services scheduler is now just a nova scheduler that can interact with storage and network component
to make decisions, but not a universal scheduler that can be used by other components. Did I make it right?

Looking forward to hear from you.

Thanks and regards!


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20140305/9514bfae/attachment.html>

More information about the OpenStack-dev mailing list