Tip:
Highlight text to annotate it
X
Now in this Overview we just give you some
examples of the current projects, I think we said
there were 318 of those so far, and... they're just
selected in a not terribly scientific fashion to be...
[pause]
cover a reasonable range. An earlier...
project developed a hybrid version of Hadoop,
or more precisely, of MapReduce... to do a
privacy preserving gene read mapping
where they developed MapReduce with...
[pause]
a small core system which could be run locally,
which did some transformation on the data
so that most of the computing could be done in...
an anonymized fashion on a public cloud.
One of the key problems with human gene processing
is the privacy requirements, and, of course,
public clouds... although they try very hard,
they intrinsically have less privacy than
a private system where nobody else is using the environment.
Then we have... Project 132, which won
the IEEE Scaling Challenge at CCGrid 2012.
And that was a project on power grid sensor analytics,
with a distributed Hadoop. So notice those
first two actually had... were using MapReduce.
156 is a project looking at high performance
networking using the machines in a specialized
fashion to develop Infiniband over Ethernet.
And that's actually pretty interesting for FutureGrid
cuz this is the type of model we expect
FutureGrid to be using. An interesting project,
172, transactional memory on clouds,
was thinking of fundamental computer science
issues of distributed concurrency control,
sometimes called software transactional memory,
and that was submitted for a relatively
prestigious international conference in this... field.
One more page of examples, slightly different type.
Those are mainly... core computer science.
Here we have... couple of project, 42 and 45,
which were developing so-called pilot job
abstractions and its... their application using SAGA,
which is a well-known OGF standard, and that is
developing the cyberinfrastructure... for XSEDE on clouds.
#130 was one on optimizing scientific workflow
on clouds. Workflow is a very important technology
largely developed for grids, but extremely useful
on clouds, cuz clouds, you want... support
the Software as a Service model, and workflows
integrate different services together.
So that used Eucalyptus on FutureGrid,
as a sort of illustration of the richness
of the applications supported.
133 was a supply chain network simulator
which supported Monte Carlo simulations
in that area of logistics. 257 was looking at
particle physics data analysis for the well-known ATLAS
Large Hadron Collider experiment, and it used
a mix of FutureGrid and Canadian cloud resources
to study data analysis with up to 600
simultaneous jobs, and that used both Nimbus
and OpenStack virtual machine managers.
254 is an interesting project, still ongoing,
in that it is looking at different approaches to
storing and querying online social network data,
and that's an example of one that's using these
large disk machines, it's actually running on
Delta and Bravo, because it's looking at these...
[pause]
distributed database, MongoDB, Riak, and HBase.
HBase uses HDFS in its background
and the others have dedicated... disk support.
And those are all examples of modern cloud
environments being used to support academic
social networking resource. 323 was a recent
project which was looking at SSD solid state disks
and performance benchmarking them for, actually,
HDFS, the MapReduce technology
on the San Diego supercomputer system
called Lima, which I mentioned earlier,
specialized machine having SSDs.
[pause]
The next slide is pretty important, cuz it describes
a very... important use of FutureGrid,
which actually is interesting cuz we didn't really
stress it in our proposal, but it... just is
one of the things you discover that there's a lot of
interest in education and training, because
FutureGrid offers the type of rich environment,
the ability to dynamically change the environment...
and that's extremely attractive. We've had 28
semester-long classes, a total of over 563 students,
and those have covered cloud computing,
distributed systems, classic computer science course,
scientific computing, and also a data analytics class.
We've had three one-week summer schools,
and those had a total of... those have actually
been the largest individual projects,
with a total of 390 students.
Two of those summer schools were particularly large.
One was a small summer school offered...
but a very important one... offered to a set of
historically black colleges and universities
to introduce them to clouds and... Hadoop.
We've had 7 smaller workshops and tutorials,
1- to 3-day. The last one of those was Indonesia
with some forty students involved, and we've also
had those at conferences like Supercomputing.
We also have supported several undergraduate
research projects, and all of this comes from
around twenty institutions. We're moving to add...
massive online open courses to FutureGrid,
cuz FutureGrid seems to naturally support that model,
and we're sort of showing this by actually adding
MOOCs to describe FutureGrid. And this
particular MOOC is one of those...
developed for this purpose. There are a couple of
links here. One is the set of MOOCs corresponding
to our last summer school a year ago,
and a new MOOC that's just been produced by
Renato Figueiredo on his... peer-to-peer
technology named ViNE, virtual networks.
And an important aspect of FutureGrid is that
we not only support... in general education and
training, we also have the specialized appliances.
Appliances are poorly configured images,
and they support important technology such as
Condor, MPI, Hadoop, and iterative MapReduce,
the latter all on... matter of fact, all of those would
be on virtual clusters supported on FutureGrid.
[pause]
Here's a little expansion of the support of
classes on FutureGrid. We've optimized our
portal to support classes, so you can set up
a class and then add students conveniently.
To have a class, you make a project proposal,
which can be a semester class, a workshop,
or what have you. It's then approved as a
FutureGrid project to become active,
and then you can add those users to the projects.
And users create accounts, and project leaders,
which is typically the teacher or somebody,
or maybe the teaching assistant, and they can
authorize the students to gain access to resources.
And then the students can, of course,
interactively use the FutureGrid resources.
[pause]
Note that we've had some trouble in the past,
in that the current... actually the past technology
was rather hard to use, with not terribly good
interactive interfaces. But now with Phantom
on Nimbus and Horizon on OpenStack,
replacing command line... systems like
eucatools, then I believe the open source clouds
have begun to... becoming easier to use.