Stop buying more gear than you need

Part of the fun of being a consultant is having a ringside seat at vendor meetings, generally as an advocate for the client, otherwise known as the vendor’s sales prospect. These meetings are simply chock-full of unintentional humor, usually presented by a straight-faced salesperson who honestly doesn’t know any better. Few and far between are the salespeople who understand their product’s true strengths and weaknesses. Most just read their own PR and sell, sell, sell.

This is how just about every company spends too much on products they don’t need. Fact is, when neither the salesmen nor the client actually understands their requirements, magic happens. Ten times out of ten, that magic is bad mojo for the customer. I have been privy to many examples of this, but one instance in particular stands out. The names and key facts have been changed to protect the innocent and, unintentionally, the guilty.

It was a relatively simple virtualization deal — a single-blade chassis with some number of blades and a SAN back end was required to virtualize about 30 light- to medium-duty servers, with the potential to grow. An RFP was distributed, and several proposals were selected for in-person meetings. The first vendor was a platinum reseller for Brand X and had packaged everything in a nice, neat bow. The solution started with eight dual-CPU quad-core blades, 64GB of RAM, and an 8Gb fibre-channel SAN back end for a total cost of well over $250,000. It worked out to nearly $10,000 per planned virtual server.

When I mentioned this to the vendor in the meeting, I was suddenly awash in completely false claims: “Well, VMware won’t run on anything other than 8Gb fiber-channel,” and “Our testing shows that you can get maybe four or five VMs per 8-core blade.” The CFO was blown out of the water. The reality is that the proposed solution could have run the entire existing datacenter probably three times over, and the vendor was trying to sell it as if it was barely enough. I can only imagine how often that pitch works when dealing with the uninitiated.

The next vendor was a bit more realistic, coming in at $150,000 with a 4Gbps FC SAN and slightly lower-spec blades. It was still way, way more than was necessary, with the vendor’s worst sales offense coming in the guise of dual-path FC licensing that couldn’t even be used with VMware, when the vendor knew that this was for VMware exclusively. The rep was trying to sell that on top of the original deal for an additional $35,000.

I’ll spare you the rest of the sordid details and tell you how it all worked out.

That infrastructure is currently running around 75 light- to heavy-duty VMs on a single-blade chassis with four blades. Each blade runs two 2.66GHz Intel 5355-series CPUs, with 16GB of RAM, and uses VMware’s software iSCSI initiators to a single-shelf SATA-based iSCSI array. There are no performance problems, and indeed, the ESX farm consistently runs 40 percent RAM utilization, averages around 20 percent CPU utilization (with spikes, naturally), and has plenty of room to grow. The total cost of hardware was around $70,000.

There’s a counter-argument to be made here — that underbuying can sabotage a new initiative. That’s true: Buying that particular solution with dual-core CPUs would have saved a bundle, but those blades would be at capacity already. That’s why you need to hire someone who knows what they’re doing to sit in the meeting with you as your advocate. Spend a few hundred or thousand dollars on that expert, and they’ll save you 100 times their fees quite easily.

I’ve used a virtualization example, but it’s the same everywhere. One rule of thumb is that when a vendor specs an item that’s state-of-the-art (like 8Gbps Fibre Channel) and you’re not sure you need it, chances are that you don’t. The people who need 8Gbps FC not only already know, they’ve been testing it for a while now. Heck, I’ve had 2,500 active-port networks that don’t push much more than 1Gbit between core switches at any given time.

The fact of the matter is that unless your infrastructure is constantly bumping into a hard performance ceiling, you can save huge amounts of money by carefully inspecting your needs. I know this sounds simple, but time and time again, I see money wasted on hideously over-spec’d equipment. The cost differential between 10G and 1G uplinks on switches is substantial. If you have trending software running (and you do, don’t you?), you might be surprised to find that you rarely — if ever — reach a full gigabit on those uplinks. If you do need 10G, you’ll probably already know, since you’ve been having problems with 2Gb or 4Gb bonded links.

Remember: Most salespeople simply regurgitate the party line and stay as far away from reality as suits their commissions. Want to trim your budget and save money? Then start with sales meetings — and make sure someone who knows what you really need is on your side of the table.

This story, “Stop buying more gear than you need,” was originally published at InfoWorld.com. Follow the latest developments in hardware on InfoWorld.com.