tag:blogger.com,1999:blog-7006618857347978140.post2925959519903328120..comments2009-05-09T19:34:39.769-07:00Kevin Lawtonhttp://www.blogger.com/profile/03442192017196947120noreply@blogger.comBlogger1125tag:blogger.com,1999:blog-7006618857347978140.post-1080401669308966312009-05-09T19:34:00.000-07:002009-05-09T19:34:00.000-07:00I agree, and this is an interesting new way to attack the problem. <br /><br />Arguably, underutilization due to excess "head-room" is a huge problem for physical servers as well as for VM's. Since the latency for load balancing physical servers (e.g. bringing up a physical server from standby) is at least as bad as the latency for VM migrations now (e.g. using "virtualization 1.0" technology), your proposed methods will become more advantageous over time, not less. In reality, migrating/re-provisioning physical servers is an order of magnitude slower than for VM's. <br /><br />We have been researching the power efficiency gains of different storage topologies/architectures and agree that this sort of intelligent storage sync (if you'll forgive the oversimplification) could have huge power efficiency benefits. What is especially interesting is that with physical servers, this approach to reducing head-room only really works for homogeneous servers (e.g. server pools). For the case of large numbers of VM's, it can be heterogeneous, with lots of different server applications, perhaps only with a common underlying OS. For example, we see a case in our labs where the active, unique memory footprint of a RHEL server is often only 12-25 MB. (!)<br /><br />This is one more driver towards virtualization (away from physical servers), and towards novel VM management technologies.Toddhttp://www.blogger.com/profile/15595087554111580752noreply@blogger.com