How Big Is Your FRU?

The Sun Modular Datacenter S20 is the world's first virtualized datacenter built into a shipping container and optimized to deliver extreme energy, space, and performance efficiencies.

Earlier this week, Sun Microsystems announced that its Project Blackbox was now a commercially shipping product. I have to confess that when they first told me about this effort I saw it as a nice showcase innovation — something they could use to demonstrate how densely racks could be configured and how energy efficient their products were. They could drive it from city to city for in-person demonstrations. Nice marketing idea. But I didn’t see the practicality to real enterprise data centers. Who’d be willing to buy a container and park it outside their data center? Yeah, that’s secure.

But then Rackable came by to show us their container.

Again I saw the “park it outside” example, but they also talked of data centers being designed with indoor loading docks pre-wired to accept data center containers. Then HP came by with containers for disaster recovery purposes.

Then last week at the Cisco Nexus 7000 announcement, Debra Chrapaty, Microsoft’s corporate VP of global foundation services, stood up and told the audience, “Think containers. They’re for real.”

Turns out she’s buying them and she’s definitely not alone. The Stanford Linear Accelerator Center is using them to rapidly expand their high performance computing capacity without having to make physical plant improvements in their existing compute building.

The key to understanding the container value is in how the concept of the field replaceable unit (FRU) is defined by the infrastructure and operations professional. We usually think of hot-swappable drives, NICs, and power supplies as FRUs; and they are if your unit of “element to be managed” is a server.

In the case of Microsoft, which manages hundreds of thousands of servers across myriad data centers and colocation facilities, the unit of measure can’t be the individual server; it has to be rack, the service, the facility. Google has been well publicized for building its own servers for its mega-sized data centers (megacenters). They aren’t building robust servers; they’re stripping them down because they have no intention of fixing the server if a component fails. It’s far more economical for them to throw out the whole thing. The server is the FRU.

But for the world’s megacenters and the operations professionals who run them, the server as the FRU may be becoming old thinking. The data center container makes the case that in fact the FRU may just be a lot larger than that. Imagine if your ERP system ran in container 0015 and failed over to container 0034. Is the next generation data center rolling down I-95 right now?

By James Staten

Check out James' research.



re: How Big Is Your FRU?

I wrote a quick response to James's article at the Cisco Data Center blog@ sure how the Viagra ad got posted here, may be worth deleting it :)

re: How Big Is Your FRU?

The issues to consider with containers is that while they allow for dense deployment of servers, they represent a solution for the least constrained aspect of todays data centers, i.e. floor space.Containers still require power and cooling plants to run - not a small obstacle. If you already have underutilized capacity in the transformation gear, switch gear, breakers, wire gauge, UPS, and generator, etc on the power side, and over sized water delivery capacity, chillers, and heat exchangers on the cooling side, bring on the containers if that fits the requirement. However, the biggest expenses with uplifting an existing data center is replacing the power and cooling infrastructure as well as the delivery system including the investment in a copper mines worth of the material needed.If the issue is needing more floorspace, most data centers lacking floorspace are highly unlikely to have any available power and cooling to spare.