How to run your business on four rack units

For the past eight months or so, I’ve been running a de facto virtualization test in the lab. I took a variety of lab boxes and internal servers, virtualized them on VMware vSphere 4.1, and ran them all on two physical servers. It may not seem impressive, but it indicates what’s possible for small to medium-size corporate computing. And for those looking to consolidate, it’s highly relevant.

These are not high-powered boxes. In fact, both servers fit into a single rack unit, using Tyan’s YR190-B8228 side-by-side dual-socket server. Within that single rack unit, there are a total of four physical processors: two AMD 4162 EE CPUs with six cores apiece running at 1.7GHz, and two AMD 4170 HE CPUs in the other half of the box, with six cores each running at 2.1GHz. With 32GB of RAM in each side, that’s enough horsepower for a surprising array of VMs and services.

[ Read Paul Venezia’s virtualization shoot-out comparing products from Citrix, Microsoft, Red Hat, and VMware. | See Matt Prigge’s “VMware snapshots: The good and the bad.” | Subscribe to InfoWorld’s Virtualization Report newsletter. ]

The storage is a Synology RS3411RPxs: a 10-spindle SATA array with redundant power and multiple 1G and 10G links, though only the 1G interfaces are in use for this particular experiment. There’s a QNAP TS-EC1279U-RP in play to address other tasks, but it also contains backups of the whole shebang, replicated every few hours.

Each physical server has two gigabit links to storage and a single gigabit link to the front-end network — small potatoes indeed. However, the services running on this cluster handle an equivalent load to a moderately sized business (50 to 100 people). Naturally, this is an estimated workload, but these boxes get pushed hard during the day and continue on with various processes in the evening.

There are domain controllers, the vSphere vCenter server, internal mail servers, external mail relays, file servers, backup servers, application servers, database servers, and even a PBX server running Asterisk that handles all the phones, both analog and SIP. It’s not tasked with highly transactional databases hit by thousands of queries a second, but it’s a relatively busy little data center in a box. Well, three boxes — including the switch, it takes up all of four rack units.

In looking at the statistics over time, it’s clear there are no bottlenecks anywhere for the load past and present. During normal operation, the CPUs on each server hover around 10 to 15 percent utilization with spikes to 40 percent or so when one VM gets a little more action. RAM utilization averages out around 25 percent per box; if one unit fails, there’s easily room for all those VMs to restart on the other side. VMware’s DRS handles the load balancing, and HA handles any hardware failure event. In fact, the only thing that could be added might be more RAM if the VM count increases to any substantial degree.

Looking at this data center microcosm, we see I’m easily running a varied workload on low-power, low-cost chips, servers, and storage, and I have plenty of room to grow. If anything, this is where smaller infrastructures need to be concentrating.

When planning and specifying servers for new virtualization projects, it’s enticing to go with the biggest, baddest, fastest CPUs in an effort to “future-proof” the resulting infrastructure. The theory is that if you get the newest, fastest chips available, they’ll have a longer usable lifespan than if you go with cheaper, slower CPUs. However, the returns on that investment may not always be what you think.

The fact is a significant number of virtualized infrastructures use far less CPU resources than anything else. They chew up RAM, but RAM is cheap and easily added. Spending money on RAM and faster disk is almost always going to provide better overall results than dropping several thousand dollars per CPU on the fastest chips on the market.

There’s always a question of redundancy as well. Having only two servers running an entire company’s server infrastructure may seem precarious to many, but the reality is that with the load-balancing and high-availability features in modern virtualization solutions, both servers would have to fail at the same time to actually take down everything. If it’s that much of a concern, you can always add one or two servers later on instead of just increasing the RAM in the existing boxes. The primary storage in this case may not have redundant controllers, but it does have redundant power supplies and is easily backed up to the QNAP box through the use of simple rsync.

This solution isn’t tuned or spec’d for anything beyond normal small- to medium-sized-business computing needs. However, the fact that the cost of all of these components is well under $20,000 and can support a significant number of users is certainly food for thought.

This story, “How to run your business on four rack units,” was originally published at InfoWorld.com. Read more of Paul Venezia’s The Deep End blog at InfoWorld.com. For the latest business technology news, follow InfoWorld.com on Twitter.