Hi all,
My MRQL talk at ApacheCon was well-attended (about 40 people). There
were many questions after the talk and, later, after the session. One
said that we should focus on the language description and let the other
projects provide their own implementation of the language. His concern
was that MRQL developers cannot have a full grasp of all capabilities of
all these platforms, so they will not do as a good job as the developers
of the platforms. I explained to him that, since we do not have a large
community, it would work better for us if we provide a prototype
implementation for various platforms and then let the developers of the
platforms refine and optimize them. Also, I talked with Fabian from
Flink about the possibility of adding Flink Streaming to MRQL. It seems
that Flink Streaming is still work in progress. I finally met with Lee
Moon Soo. His talk on Zeppelin was very interesting. He doesn't use MRQL
any more but he is working on supporting Flink (Spark already has it's
own data visualization system). There was a lot of interest in Spark.
There is even a full-day dedicated to Spark today, which I couldn't
attend. Many developers believe that Spark is replacing map-reduce. I
think this is great but I am very puzzled. Spark is based on high-order
functional programming and most developers are unfamiliar with
functional programming or even Scala. I think they are just impressed
with Spark performance, but then again there are alternatives (eg,
Flink) that are faster. I also had a discussion with Marvin (the IPMC
chair). It was very nice of him to approach me and ask me about the
project. I explained to him the problem that prevents us from graduating
to TLP, namely community growth, especially developers. He said that not
all projects are expected to have as big community as Spark, etc; there
are small projects too and IPMC supports them too. He said that the best
way to recruit developers is to spread the word about MRQL, so that
people will use it, companies will adopt it, users will request changes,
then they will submit patches, and become contributors, then developers.
This takes a lot of time for some projects. He said some projects, such
as Mesos & Aurora (there was a talk earlier about this), had no releases
and no new developers for about a year, but eventually they increased
their adoption and community, and graduated. I asked him about
monitoring downloads & web site hits. It seems that we cannot count
downloads but we can monitor web site traffic using google web site
analytics. I will look into it. In summary, this was a very interesting
conference: many interesting talks (some about project
incubation/graduation which are helpful to us), and a chance to meet and
talk with other developers.
Leonidas
On 3/30/15 10:51 AM, Leonidas Fegaras wrote:
Thanks Alan,
I think Apache Storm is a good choice: it's very popular and can be
nicely integrated with MRQL streaming. I will put this on the slides.
Leonidas
On 3/30/15 9:51 AM, Alan D. Cabrera wrote:
These are great slides. I wish I was going to Austin.
If it makes sense, I would explicitly list some of the distributed platforms
where support could be added. If there are any high profile projects to be
integrated with MRQL it would be good to state them, imo, to attract some eager
developers. :)
Anyway, great slide deck!
Regards,
Alan
On Mar 29, 2015, at 9:52 AM, Leonidas Fegaras <[email protected]> wrote:
Hello,
I will give a talk on MRQL at ApacheCon15 in Austin.
My goal is to spread the word on our project, expand the user community, and
recruit developers.
The slides are available at:
http://lambda.uta.edu/mrql-apachecom15.pdf
Please let me know if you have any corrections or additions.
Thanks,
Leonidas Fegaras
.
.
