Companies with critical Web applications need to make decisions about how much they're willing to invest to ensure those applications are available to customers and employees around the clock. One way to achieve this type of Web availability is to measure it by a system called "five nines," which has been around the computer industry for quite some time, primarily to assess hardware uptime. Aaron Goldberg, vice president and principal analyst for Ziff-Davis Media, has recently done some research into the area of applying "five-nines" to Web site availability.
What is "five-nines"?
It's a metric that speaks about system availability by the number of nines. One nine is 90% uptime, two nines are 99%, three nines are 99.9% and so on. It's a desire to have essentially no unplanned downtime; it's a commitment that says a company is going to the extremes of availability.
What are the factors in the decision about what level of availability is right for your site?
You have to determine the sensitivity of the data; how important availability is to customers, employees and others; the impact on the underlying database (if the system goes down in the middle of a transaction and the back-end database is corrupted); and competitive issues like the costs of the system not being available. After taking all this into account, you come up with a number for each application -- a phone system needs five nines; online banking may need three or four; an HR application that tracks employees' time off may need just one nine.
After you figure out what availability each application needs, what are the next steps?
You have to look at your overall architecture and see where the applications are hosted -- the types of hardware and software. If you have four applications on one server, then all four applications will have to have the same level of availability -- you can't make one application have three nines and another have four if they're both living on the same physical machine.
How difficult is it to manage and track different levels of availability for different systems?
Some people choose to outsource the hosting of their highest-availability applications to an ASP or MSP (managed service provider). They can then group all the lower-availability applications on their in-house servers. But you can't make different in-house applications wildly different, unless you're willing to invest in separate servers, hot-spare standbys, UPSes and so on.
How do costs play in the equation? Is it safe to assume that it gets more expensive as the number of nines goes up?
It generally gets more expensive, depending on your environment. But going from four to five nines could be a 10% or a 400% increase. It depends on what you've got.
How do you figure out "failure acceptability" -- that it's okay for an application to be down for a particular amount of time? Who needs to be part of that decision?
You need to look at percentage of downtime versus uptime and how long the downtime lasts. EBay may have 100 outages/month for a half-minute and no one notices. But if they have 10 outages for 30 minutes each, people get pretty ticked off. As for who's involved in the decision, IT absolutely has to have business management buy-in with this. IT may think an application is a three, but business executives may rate it a four.
Do you have a sense of what people are actually doing when it comes to Web-site availability? Are they investing in five-nines for everything, or just for some things?
Sites with brochure-ware -- those that are mostly doing marketing or providing information -- are generally at one or two nines. Those doing minimal e-commerce are at two or three nines. Heavy e-commerce sites are between three and five nines. Thoughtful IT executives are trying to be pro-active and are asking how many nines they really need, and what the business is really willing to pay to achieve that level of availability.
About the author:
Johanna Ambrosio is a freelance writer in Marlborough, MA. Reach her at firstname.lastname@example.org.