We are using a Windows (OS) based laptop having VMware installed on it. We need to develop a Grid, out of lets say, three virtual computers/machines created by VMware (on this laptop).
Hello, Mr Puri,
Wish You happy and successful new year!
Yes, I did the same - Grid, consisting of three virtual Debian machines, installed on a VMWARE on my laptop. For managing the GRID I used the Globus toolkit. Here is a good quick-start installation manual (http://www.globus.org/toolkit/docs/4.0/admin/docbook/quickstart.html).
The main problems, as could be expected, were that such a GRID environment is resource-consuming. As my laptop is not very powerful (Dual Core 1.6GHz, 1GB RAM), the GRID worked slowly.
Hello Mr. Penev,
Wishing you a very happy new year - 2013! Hope this year brings you - true fulfillment, and success in your pursuit for knowledge!
Thank you for writing and submitting your reply to me at "Research Gate". I appreciate it. I will try to look and work on a solution based on your inputs.
Besides the above, I just wanted to check if you would like to guide me on the next step (matter mentioned below) which is important for my project -
"Machine Learning" has to be implemented (using - "Artificial Neural Networks, Genetic Algorithms and Fuzzy Logic") on this Grid, to - "Predict the Time and Resources Consumed by Applications running on these three virtual systems." Modern machine learning techniques able to handle large number of attributes should be used, taking into account application and system specific attributes (e.g., CPU micro architecture, size and speed of memory and storage, input data characteristics and input parameters). Also, state the most suitable machine learning technique(s) for predicting spatio temporal (space and time) utilization of resources by applications and systems.
Regards,
Parikshit
Parikshit,
The project seems very interesting!
I have been working on predicting the execution times of jobs with known input parameters.
The environment, however, is limited to identical connected computers. I have proposed a genetic algorithm to schedule the jobs in a cluster (or GRID), consisting of identical computers, using the predicted jobs' execution times.
Your project is more complex, and of course, more realistic.
Regards,
Ivaylo
Thank you for the above comment Ivaylo. Let me check what I can do and I may get back to you with further comments. Have a great week ahead.
Good day.
The main problem I see with using virtual machines such as VMWare is that each virtual machine must reserve it's own RAM space and build its own process tree managed by a separate scheduler. As you have noted, this may be a bit too much of a load for a single dual-core CPU.
I think you would get better performance (at least one order better) if you could use Debian as the main operating system on your hardware and spawn your virtual machines as chroot processes - this is known as User-mode Linux (http://en.wikipedia.org/wiki/User-mode_Linux). Each virtual machines runs as a separate process on the host. Context-switching is greatly accelerated, and RAM provisioning is more efficient since each machine is assigned just the amount of memory it really uses.
PS: I cite Debian since you already have experience with this distribution, UML does not depend on any particular distro.
Just my 5 cents...
-Alan
Thank you Mr. Alan, for your input. I'll get back to you if I need your further guidance on the concerned matter.
- Badges
- Science topic
- Similar topics
- Computer Science