Playing hardball with your hoster

The most expensive real estate you'll ever lease is collocation space. A single rack in a cabinet, roughly two feet square, runs about $1,000 per month, and that's before you start paying for bandwidth or managed services.

Before you sign on with a Web hoster, here's what you need to do.

Step 1: Analyze your needs

Here are the questions you need answered:

1. How many rack units and amps will each piece of equipment consume?

2. In terms of management, will you install your own KVM switches and terminal servers, or will you be paying for managed services?

3. How much bandwidth will you need to run your applications?

4. What type of monitoring tools do you need?

Step 2: Create an RFP

In addition to those four questions, here are some other items you need to consider when developing a request for proposals:

Network - What committed information rate (CIR) are you signing up for, and at what cost per megabit? What are your burst costs should you go over your CIR, and can you be back billed at your CIR rate for long-term burst overages?

Power - Ask for circuit utilization monitoring (you don't want to pop a breaker), and ask about power guarantees often. Despite the presence of power back-up systems, contracts will only provide a best-effort clause.

HVAC - What guarantees are offered for environmental conditions? Most HVAC systems require a persistent source of water. What redundant sources are in place?

Fire - What system is in place for putting out a fire, and what impact might it have on your equipment?

Security - What's in place to control access to your equipment, and to ensure 24-7 access? Can you get third-party vendors in to service or repair your equipment?

Remote hands - Sometimes you just need a tape changed or the power cycled on a piece of equipment. Is there a cost? What response time can you get?

Receiving - Will the collocation facility accept delivery of equipment on your behalf? What protections are in place?

Maintenance windows - Some vendors don't guarantee availability of the network during maintenance windows. Check it out.

Managed services - What services are you considering, and what guarantees are provided for them? For example, should you opt for the collocation facility to provide data backup? What guarantees are made? What restore windows are offered? What off-site storage plan is offered?

Contract termination - What conditions let you cancel a contract: bankruptcy, change of ownership or degraded level of service?

Put all this information into an RFP that you will supply to each prospective vendor. Then get ready for face-to-face meetings with each vendor.

Step 3: Prepare for negotiations

Expect to be able to answer the following questions from each vendor and understand what they are really asking:

1. What's your budget for this project and is the funding approved? (How much money can I make from you and is this just a pipe dream?)

2. What's your time frame? (Are you going to help me make my quota this month/quarter?)

3. Who has final approval for this project? (Am I speaking to the decision maker?)

4. Who else are you talking to? (Who's my competition?)

5. What are your decision factors? (What should I be pitching?)

Use care in how you answer these questions. The wrong answer on any question can shift the power balance in negotiating your final costs and service-level agreements (SLA).

Step 4: Negotiating the SLA

SLAs are guarantees of a specific quality of service for a particular feature. An SLA should include a clear statement of what feature or service is being provided, how the quality of that feature is measured, what degradation of quality is considered a trigger for invoking the SLA, how that trigger is initiated and what credits are provided in response to the trigger.

Here's wording that we found in an actual SLA:

"Network availability is a derived, percentage-based measurement based on the number of one-hour periods of 100% packet loss within a calendar month. If the network availability over our chosen routes falls below our guaranteed levels, all eligible customers will receive one day of credit for the affected service on their next monthly invoice.

"We proactively monitor, notify and credit your account if there is an infraction - so you won't have to catch us in the act of breaking an SLA threshold. If a service point outage is longer than 30 minutes, we'll also credit your monthly bill."

If you read this thoroughly, you might conclude that 59 minutes of 100% packet loss gets you nothing. You would be right. You might also conclude that 59 minutes of 100% packet loss occurring multiple times per month would get you nothing. You would be right again.

One positive element is that you don't need to initiate the trigger yourself, a common requirement in many SLAs. Often you need to open a trouble ticket to report the degradation of quality for an element of service yourself. If you're not doing your own monitoring, this can be problematic. Given the low cost of setting up your own basic monitoring, and the high cost of downtime, you should strongly consider some level of in-house monitoring.

Tighten it up

There are many ways that this element of an SLA could be tightened. For example, because bandwidth billing is done on 5-minute intervals, consider it appropriate to have availability measured in the same increments. An outage of 100% packet loss of 5 minutes could be a trigger.

You also might want to consider a lower threshold than 100% packet loss. Losing 50% of the packets heading to your e-commerce site would be devastating because under the SLA in question this could go on for hours without triggering the availability element.

Other common network SLAs cover latency or packet loss less than 100%. Again, read the fine print for each of these and understand exactly what metrics are being used to measure these elements, over what time period and what degradation of quality is required to trigger a credit.

Does the punishment fit the crime?

Consider the credits offered in an SLA. Are they adequate to compensate you for the loss of service (usually not), and are they generous enough to provide strong incentive for the collocation partner to ensure that they are never triggered?

Push as hard as possible for meaningful credits, but understand that unless you're a very large customer, you're unlikely to see these change. Carefully calculate the maximum credit you might receive for a given incident. Often you'll find you're capped at 25% of your monthly fee, regardless of the triggers that occur. At a minimum, look to have any cap removed. Beyond that, consider your costs of downtime, and look for credits that reflect those costs.

Related Links

Apply for your free subscription to Network World. Click here. Or get Network World delivered in PDF each week.

Request a reprint or permission to use this article.