> I'm quite excited to get it working.
Great! We'll continue to work with you on this. This is definitely
something that we can make work. It may take a week or two, as we
check out all the details. Kapil is intending to try out Firefox under VNC
this weekend, so he can see the same thing that you're seeing.
Our current guess is that the issue maybe is _not_ epoll. DMTCP
currently disables epoll since it can't handle epoll properly. It then
issues the warning that you saw. But in the past, we've seen Firefox survive
that and checkpoint/restart correctly anyway.
We'll keep you informed.
Best,
- Gene
On Fri, Feb 24, 2012 at 04:45:05PM -0500, Jason wrote:
> Hi Gene,
>
> I have been using a modified copy of Firefox 3.6 so if it can be configured
> to not use those methods I will definitely recompile it and see if it works.
> Regarding running it headless, we do interact with the X window so that would
> not be possible for us. When I tried it out earlier this week, checkpointing
> uncompressed to ramdisk with DMTCP was very fast indeed, definitely much
> faster than my current solution with OpenVZ. Even though I couldn't quite
> launch the browser, I did get a good sense of how much faster it will be. I'm
> quite excited to get it working.
>
> Have a great weekend,
>
> Jason Kay
>
> On 2012-02-24, at 4:15 PM, Gene Cooperman wrote:
>
> > Hi Jason,
> > [ I'm continuing the public thread on dmtcp-forum, in case some of this
> > information is useful to others. I hope that's okay.]
> >
> > I just got off the phone with Rohan. He tells me that he expects to have
> > working code by the end of this Tuesday. So, here's the story:
> > 1. We can probably give you a new tarball on Wed. that works for epoll
> > 2. From my quick look, it seems like Firefox can also be configured
> > to use inotify in addition to epoll. DMTCP doesn't yet support
> > inotify. That could be another week's work.
> > 3. I think that Firefox can be confrigured to not use epoll or
> > inotify. Does this help you?
> > (Kapil, do you remember how to do this?)
> > 4. As for the speed and size of checkpointing, in DMTCP, try turning
> > off gzip compression and checkpointing to a RAM disk. The size
> > will be larger, but it will run _very_ fast. There are also newer
> > compression utilities that are better than gzip in terms of time
> > and compression.
> > 5. The VNC trick should work with graphics, and it's possible to use
> > just the core Firefox library without graphics in certain special
> > situations.
> >
> > Best,
> > - Gene
> >
> > On Fri, Feb 24, 2012 at 02:24:27PM -0500, Jason wrote:
> >> Hi Gene,
> >>
> >> Thanks for your response, I'm looking forward to assisting in any way I
> >> can. I'm developing an application integration platform which uses Firefox
> >> and other browsers in the cloud to interact with legacy and third party
> >> websites. This allows developers to write APIs which integrate websites
> >> they don't have full control over.
> >>
> >> The platform has many similarities with the popular Kapow Katalyst
> >> platform but is much more code oriented, and features a browser based
> >> debugging environment which allows the end user to write robots in a
> >> highly efficient manner due to its checkpointing capability. We currently
> >> use OpenVZ to provide checkpointing. Currently this creates dumps of 200mb
> >> in size or more and the process is IO bound, creating a noticeable delay.
> >> We also create dumps any time an unexpected error occurs so the operator
> >> can debug the session. Our platform uses the open-source WebDriver API
> >> from Google, and I think there will be a lot of interest from that
> >> community of developers in your project.
> >>
> >> I was able to get your program working with tightvnc and openbox, but when
> >> I tried running Firefox I got a blank window, and figure it was probably
> >> because of the epoll errors. I'm glad to hear that Rohan Garg will be
> >> taking this on. Let me know if I can be of any assistance in testing.
> >>
> >> Jason Kay
> >>
> >> On 2012-02-22, at 5:25 PM, Gene Cooperman wrote:
> >>
> >>> [ I forgot to cc to Rohan Garg. Here is a separate cc (not on
> >>> dmtcp-forum). Best, - Gene ]
> >>>
> >>> On Wed, Feb 22, 2012 at 12:53:25PM -0500, Gene Cooperman wrote:
> >>>> Hi Jason,
> >>>> First of all, since Firefox is graphics-oriented, you'll have to
> >>>> use the trick with VNC. See:
> >>>> http://dmtcp.sourceforge.net/supportedApps.html
> >>>> for details, and let us know if you have any trouble with that.
> >>>> Second, Firefox can be built with or without one of three newer Linux
> >>>> system calls (introduced about 5 years ago). Your output is showing that
> >>>> you're hitting one of them ("epoll"). Until the last few years, many
> >>>> Firefox
> >>>> builds were configured to not use these calls. More recently, we're
> >>>> seeing
> >>>> this more.
> >>>> So, we are now extending DMTCP to support epoll/eventfd/inotify.
> >>>> The good news is that a student here (Rohan Garg) took it on as a
> >>>> short-term
> >>>> project (a kind of "warm-up" in getting to know the DMTCP internals).
> >>>> I'm taking the liberty of cc'ing to him. This way, he gets credit, and
> >>>> I'm sure he'll be pleased to see that there are people who want to use
> >>>> this.
> >>>> So, we hope to put something in the svn soon, and we could give you
> >>>> a tarball even sooner, if you should want to follow up jointly with us.
> >>>>
> >>>> By the way, I'm just curious, but what is your requirement for
> >>>> checkpointing Firefox. We had previously given this a lower priority
> >>>> because it seemed like checkpointing Firefox was "a nice hack", but
> >>>> of lesser utility. If there's an important use case, we'll make sure
> >>>> it's closer to the center of our radar. :-)
> >>>>
> >>>> Thanks,
> >>>> - Gene
> >>>>
> >>>>
> >>>> ------------------------------------------------------------------------------
> >>>> Virtualization & Cloud Management Using Capacity Planning
> >>>> Cloud computing makes use of virtualization - but cloud computing
> >>>> also focuses on allowing computing to be delivered as a service.
> >>>> http://www.accelacomm.com/jaw/sfnl/114/51521223/
> >>>> _______________________________________________
> >>>> Dmtcp-forum mailing list
> >>>> [email protected]
> >>>> https://lists.sourceforge.net/lists/listinfo/dmtcp-forum
> >>
>
------------------------------------------------------------------------------
Virtualization & Cloud Management Using Capacity Planning
Cloud computing makes use of virtualization - but cloud computing
also focuses on allowing computing to be delivered as a service.
http://www.accelacomm.com/jaw/sfnl/114/51521223/
_______________________________________________
Dmtcp-forum mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/dmtcp-forum
