1 Reply Latest reply on Dec 21, 2007 6:50 PM by pauljacobhall

    What are the emerging trends in compute models?



      Welcome to the site.  What do you see as an emerging trend in compute models?  Are you seeing trends to use more server/client/network resources.  Are there any new and innovative computational ways you have been considering for solving some of your industry problems?  Do you have a solution to the way buildings are being architected to handle weight/heat/energy efficiency of the data center? Share with us your ideas or problems and lets discuss.



        • 1. Re: What are the emerging trends in compute models?

          Welcome to the site. What do you see as an emerging trend in compute models?



          I see general compute devices in the data center being commoditized in the way storage arrays were commoditized. We like to call this future commoditized compute container a "processor array". A processor array features virtual machine support as a software feature, has a processor array cache for near file / block level caching (ex. booting OS's off centralized storage arrays hosted on the unfied fabric). The chassis is super dense and super cheap and will be managed as part of layer one. Commodity form factors are used (meaning 1U, .5U) to minimize cost and drive commoditization based on currently deployed technologies The compute devices will not have local storage except in the form of non-volatile RAM / NVRAM / Flash for the purpose of read and optional write caching. I/O connectivity to the chassis will be a consolidated virtualized I/O fabric. There will only be one to two wires per chassis and the wires are ideally externalized so individual nodes in the container / chassis can be upgraded independantly of the others. This also provides easy I/O troubleshooting without impacting all compute devices in the container. The compute devices will essentially be stateless so an OS can move from a virtual machine to a physical machine and back with zero hardware reconfiguration or dependancies being reqiured.



          This is a green concept, it allows us to provision quickly with support for moving to bare metal easily. It also improves our ability to provide machine high availability by default. The processor array design borrows from what works, uses history to evaluate future success, and lends itself to current green IT initiatives.  See attached example.



          Does this make sense? If you like it in concept - do you see areas for improvement?



          Do you see vendors doing this? If not, what is holding everyone back?