Fwd: [help] Ask for the possibility of doing a project on GENI

[Mark Berman]

Hi Jin,

Forwarding your message to the protogeni-users group for an expert opinion.

ProtoGENI folks, are rspecs or other forms of topology representation logged when a sliver is created or an Emulab experiment is swapped in? Would you (or the original experimenters) consider this data private, or could Jin get it to further his research?

Thanks,

Mark

Begin forwarded message:

From: Jin Zhu <ji...@engin.umass.edu>

Subject: Re: [help] Ask for the possibility of doing a project on GENI

Date: February 16, 2012 10:42:28 AM EST

To: GENI Help <he...@geni.net>

Cc: he...@geni.net

Hi Mark,

Yes, if it is available and the public resources, could you help me find examples of realistic topologies used by some experiments? Because I want to prove the realization of my benchmark, and test through real cases.

I use the GT-ITM tool to generate the synthetic network topology and program the mapping algorithms by c language, then use the GLPSOL to find the results of comparison.

Thanks so much.

Bests,
Jin

Quoting Mark Berman <mbe...@bbn.com>:

Hi Jin,

I'm not sure I follow what you're looking for here. Are you hoping that we might be able to help you find examples of realistic topologies used by other experimenters? It's possible that some GENI components might have such records available. I'm thinking in particular of ProtoGENI  / Emulab. I'm not sure whether that information is considered private by the other experimenters.

Meanwhile, what is your thinking on question B?

Please continue to cc he...@geni.net for a better distribution.

Thanks,

Mark

On Feb 15, 2012, at 4:10 PM, Jin Zhu wrote:

Hi Mark,

Thanks. I want to ask whether the GENI could consider give me the real virtual network requests like topologies of previous user's network?

Bests,

Jin

Quoting Mark Berman <mbe...@bbn.com>:

Hi Jin,

OK, understood. You need a network of up to 500 computers, on which you'd like to run experiments for 3-4 months. Each of the abstract topologies you map onto this network will be a configuration of 5, 10, or 20 nodes.

The next question here is essentially B on my original list of three: Once you allocate a specific topology of, say, 10 nodes, what software do you then deploy and execute on these nodes?

The reason this question is important is because it will help to determine whether you require exclusive access to the nodes or can work with shared nodes. I think it is highly unlikely that GENI can provide you with several months of exclusive access to the number of nodes you're looking for, so we should explore shared options.

Please continue to cc he...@geni.net for a better distribution.

Thanks,

Mark

On Feb 15, 2012, at 3:08 PM, Jin Zhu wrote:

Hi, Mark

The substrate network is consisted of two kinds, one with 100 nodes and the other with 500 nodes. For the virtual network requests, the size is separate into 3 kinds: 5, 10 and 20 nodes. I just consider the processing resources of nodes and bandwidth resources with links. I think the duration time of experiment is around three to four months.

I use the coordinate of topology as the input, through running GLPSOL.

Thanks,

Bests,

Jin

Quoting Mark Berman <mbe...@bbn.com>:

Hi Jin,

Great. Thanks for the info. I'm glad that I've got the right basic idea. However, you didn't answer the three questions A-C in my previous note. The answer to the question of whether GENI can provide the real resources you need will depend greatly on the answers. For example, if you want five computers for five hours, that's likely to be possible. If you want five thousand computers for five years, probably not.

Thanks,

Mark

On Feb 15, 2012, at 1:44 PM, Jin Zhu wrote:

Hi Mark,

Thanks for replying me and sorry for the delay.

Yes, you got the key point in my experiment.I use the GT-ITM tool to generate the network topology, including the substrate network and the virtual network requests. But I really need the real topology and real resources to prove the realization of my benchmark.

The input of my experiment is the virtual network requests, (including the coordinate of every nodes, edges connected between nodes, and bandwidth constraints with links, processing constraints with nodes) substrate network,(same with VN) and mapping algorithms. Then I run it through GLPSOL (an open-source linear programming solver) and compare the metics between different mapping algorithms.

Thus, could GENI offer the real virtual network topologies, bandwidth resources of links and processing resources with nodes ?

Thanks so much!

Bests,

Jin

Bests,

Jin

Quoting Mark Berman <mbe...@bbn.com>:

Hi Jin,

Thank you for your note. We're glad that you're thinking about using GENI in your work. I think that GENI may be able to support your needs, but I want to ask a few questions to make sure I understand your plan correctly.

Here's my understanding of the key elements of your experiment.

1. Your ultimate goal in this experiment is to evaluate several mapping algorithms.

2. The function of a mapping algorithm, A, is to take an abstract topology, T, and map it onto a realizable topology A(T), which is a subset of available resources, R.

3. You are already able to evaluate algorithms using several measures, but you would like to add some measures that require executing the real topology A(T) on real resources. You would like to use GENI resources for this purpose.

Assuming I've got that right (please tell me if not), then GENI ought to be able to help, depending on just what kind of resources you need in your set R of available resources. If this set of resources is something that can fit within available GENI resources, you're probably in good shape.

In order to give you good advice about what kind of GENI resources are most appropriate, we'll need to know some more about just what kinds of topologies you're looking at and what kinds of tests you want to execute on GENI resources. At first blush, ProtoGENI or PrimoGENI seem likely candidates.

So, three questions:

A. Roughly speaking, what is the set of real resources that you need? (The set R above.) I'm looking for parameters like number of computers, number and speed of network interfaces, links.

B. What software do you intend to run on these resources?

C. What is the planned number and duration of your experiments?

Thanks,

Mark

On Feb 8, 2012, at 3:32 PM, Jin Zhu wrote:

Hi?

My name is Jin, a ECE graduate student from the UMASS-Amherst. I have read the ExperimenterPort and still confused with a problem.

Now I am doing a experiment about the network virtualization. I create a benchmark to evaluate the performance of different mapping algorithms(which mapped the virtual network requests to the substrate network resources as more as possible). I evaluate it through running time, amount of successful mapping and so on. So is it available for you to provide me the real network topology with real CPU and bandwidth resources? Is it possible for me to take that resource to use? Thanks so much!

Bests,

Jin

--

Mark Berman

BBN Technologies

10 Moulton Street

Cambridge, MA   02138

(617) 873-3675

-- 
Mark Berman
BBN Technologies

10 Moulton Street

Cambridge, MA   02138

(617) 873-3675

[Robert Ricci]

We may be able to provide some anonymized topologies; Jin, I will talk
to you about this off-list.

[Leigh Stoller]

> ProtoGENI folks, are rspecs or other forms of topology representation
> logged when a sliver is created or an Emulab experiment is swapped in?

To answer this question ... we record all of the manifests in a table
called manifest_history in the geni-cm DB, at each CM. Lots of data there.

As for "classic" Emulab experiments, we save all of the NS files and the
resource mappings, but not sure how helpful those will be (will be hard to
make use of)

Lbs