1/27/2009 See comments below. -Justin ---------- Forwarded message ---------- Date: Tue, 27 Jan 2009 10:43:28 -0500 From: Renato Figueiredo To: Justin R. Davis Subject: Re: [Dmcc] {SpamScore: ssss} Re: vmware player experimenting/documentation (fwd) There are clear benefits if your workload has peaks and valleys, or is small-scale - you can pay-as-you-go by deploying appliances on services like EC2. High-end parallel jobs will not work very well yet, as far as I can tell, but throughput jobs are ok - even Condor now is working towards the ability of submitting jobs to EC2, and there are consulting firms around it that I know have been working on EC2-based appliances. I've seen people that have started studying the economics of this from the perspective of scientific kinds of workloads - from a cost perspective, purely, is it better to buy local resources or run them on a hosting infrastructure? I think in the long run the economics tendency will favor the latter - if you factor the costs of management, cooling, power, floor space, etc. --rf On Tue, Jan 27, 2009 at 10:34 AM, Justin R. Davis wrote: Any comments on Cloud-stuff? Justin ---------- Forwarded message ---------- Date: Tue, 27 Jan 2009 10:28:48 -0500 From: Jeremy Cothran To: Justin R. Davis Cc: dmcc@rcoos.org, Dan Ramage Subject: Re: [Dmcc] {SpamScore: ssss} Re: vmware player experimenting/documentation Thanks also Justin for the summary. Would also be interested in where virtualization/appliance based computing is taking place via cloud/utility type services such as EC2 and the cost-effectiveness,etc of this as compared with dedicated in-house hardware( http://developer.amazonwebservices.com/connect/message.jspa?messageID=105560 http://delicious.com/giraclarc/cloud ). Cheers Jeremy On Tue, Jan 27, 2009 at 8:06 AM, Dan Ramage wrote: Justin, Thanks for the summary. Jeremy and I basically had the same sort of thought process, how can we make our setups easier to maintain and duplicate. Dan 1/26/2009 Greetings! I guess I'll use Sam little intro to tell everyone a little about what we've been working on w.r.t. virtualization in the coastal sciences. I'll try to keep it brief and provide a few links, but if anyone has any further questions or just wanted some opinions on the how or what of virtual machines, let me know. To begin, we've been using virtualization in our research/production work for well over 5 years now. Initially we began with VMWare (before the player existed and when the server version wasn't free) but have used other virtualization products as well (e.g. Xen). Our primary reason for using virtualization is that it allows us to package all needed software things together so that users could just download a virtual machine and some monitor software (e.g. VMWare Player/Server) and go. There is no need to explain to people how to download/extract/compile/configure/etc. When we first started using virtualization techniques, we were in the process of using grid computing (aka globus-type stuff) to perform simulations. However, we found out quickly that installing/configure and keeping the software up-to-date on all-but-identical software installations was a big pain. Different versions of globus, different PERL libraries etc. resulted in numerous incompatibilities which made running anything practical very difficult. To get around this problem, we connected with a group in the UFL Advanced Computing and Information Systems Laboratory (ACIS) who have been using virtualization for grid computing. Their approach was to include all the necessary grid computing middleware pre-configured so that you just started the virtual machine and you had a grid computing node ready-to-go. The advantage here being all of the virtual machines would be identical, so all you had to do was to get your application working on one, and it would work everywhere. Through the years, the ACIS group has refined their techniques to the point where they now have what is called a "Grid Appliance" (http://www.grid-appliance.org/). Based on the grid appliance infrastructure, we've developed several coastal science applications and are working on several others. See the CI-Team and SCOOP links on the left side of the link above. We've packaged coastal models, data, archive search and retrieval tools, pre-/post-processing tools as well as web service-type visualization together making for a nice, easy-to-deploy system which is great for outreach-type activities. We're also using appliances in our IOOS storm surge testbed to provide shared access to a development testbed. jrd =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Justin R. Davis, Ph.D. davis@coastal.ufl.edu Research Associate Civil and Coastal Engineering 365 Weil Hall University of Florida Gainesville, FL 32611-6580 352-392-1436 x1528