Technologies already exist to reduce power usage
We should prevent energy usage rather than increase energy usage and then cool it. Prevention is always better than cure. We should design/architectect systems from scratch with lower power & least usage. These technologies exist today and are mature. We should not add hotter and hotter chips and more disks which require more and more cooling. This is an upward spiral in usage. Lets avoid the power usage problem in the first place. Use a low power solution in the beginning and then all following problems disappear.
We buy power hungry devices then cool them. Compounding the problem, by redesigning and upgrading additional cooling systems.
We should buy power efficient devices, that need increase in cooling - removing the problem at source, then no need to spend money on extra cooling.
Examples:
Replace all desktop PC's and laptops with thin clients. These exist from Sun and Citrix (both solutions run/support Windows apps). How many workers are really mobile. If you sit by a desk all day you do not need your own PC. Most PC's, laptops run at low CPU utilisation all day. Sunray running Unix apps and Windows apps is mature and proven existed for years. No extra staff required, security problems solved as no local data/disk/cd drives. Replacement of (2Kg) thin client takes 10mins, (20Kg) PC/laptop 1-2 days. Thin clients have no local software so no re-installs required. Many indirect savings.
Take all inactive data and put it or achive it on tape drives. Tapes require no cooling or power when inactive, disks do. This fact cannot be ignored. We can use de-duplication and then move the de-duped data blocks to tape. Use VTL's or any type of virtualisation to send data to disk. Behind the disks use tapes to store the data that is not used weekly or monthly. e.g. disk buffer pools for daily weekly used data moved to tape after a month of inactivity. All tapes are encrypted, no offsite or lost data issues, tape encryption is mature and generally available.
Virtualisation and hypervisors take extra CPU cycles and are not required to consolidate applications from many systems to one system. In the last 20yrs we have been running many apps on one shared server. Any decent OS using CPU/Mem sharing algorithms can run many apps. Most enterprise OS platforms have built in virtualisation and resource (RAM, CPU) sharing software. Large (e.g. 16 CPU, multi-core) servers can consolidate & virtualise many small servers. These large servers have reliability tecnologies far exceeding the mainframe mythological reliability. Large Unix servers and mainframes are the same thing. But more application choice and simpler staffing issues with Unix servers.
You can normally replace all web tier and small print servers, email servers with multi-core servers. E.g. approximately 5 small servers can be replaced by one Coolthreads server. With a 5 x reduction in power and cooling requirements. No need to add extra cooling/chillers in your datacenter.
Summary:
From 500W desktop PC's go to 10W thin clients
From 900W - many KW disk arrays go to 0W tape. Immense storage savings.
From many web tier 700W servers go to 400W single CPU multi-core throughput servers (e.g. Sun Coolthreads servers).
Just by using existing, proven technologies that require no extra training or new technologies we can solve todays power usage problems.
You can get this from most of the major IT suppliers. Some technologies are unique to Sun like Coolthreads servers and SunRay thin clients. I wish others could supply these too.