Availability Requirements for Network Storage

Many aspects of computing have changed since the massive buildup of the commercial Internet in the last decade. One of the most significant and painful changes has been the necessity for round-the-clock operations and the push for five-nines (or 99.999%) reliability. Most technology people would agree that living in a wired world is a great thing, but it does have its disadvantagessuch as living with a pager as a required appendage for round-the-clock administrative actions and emergency troubleshooting.

In a short period of time, American society has gone online en masse to the point where we take data access for granted. Furthermore, it's amazing how intolerant our society has become regarding system and data access. If we can't get the data we want on demandwhether it is for personal or business usewe tend to become frustrated and look for other avenues to get what we want. This impatience with data access puts an incredible amount of pressure on IT professionals.

Data access represents only one of the management vectors involved with storage today. Another issue is the hoarding habits of technology users who want to make sure they have copies of any data that might be useful in the future. Systems and networks do not always work as expected when they are continually being stretched to their capacity. While storing oodles of data online could turn out to be as natural as any other form of human behavior, it's far from natural for systems administrators to enjoy wrestling with the aftermath of it every day.

Ready access to the endless amount of Internet-resident data makes it easy to see how today's enormous disk drives can actually be filled in a relatively short period of time. When things reach their limits, something has to give. Where storage is concerned, the resolution has historically meant data access was shut off while new equipment was installed and data was transferred between devices.

It should be obvious that there can be no access to data without some sort of storage connectivity between systems and storage. What might not be as obvious is the fact that storage connectivity needs to be incredibly accurate, reliable, and scalable.

A storage system that inadvertently corrupts or loses data is worthless. Likewise, a storage connection topology that injects transmission errors and does not provide seamless capacity scaling is also a liability. There is no way to achieve anything close to five-nines availability from storage topologies that encourage disconnections as a way of life.

In order to qualify for five-nines availability, a system and its storage has to run all year, with only 5 minutes allowed for maintenance or unpleasant surprises. To illustrate how challenging this level of high availability is, the International Engineering Consortium (IEC) has published information about system availability on its website. Table 1-1, with information from the IEC website, matches availability percentages (percentage of uptime) with the maximum amount of downtime allowed to meet those percentages.

Table 1-1. Availability Percentages and the Corresponding Downtime Per Year

Availability Percentages	Downtime Equivalents PerYear
99.0% uptime (two nines)	Up to 3.7 days of downtime per year
99.9% uptime (three nines)	Up to 9 hours of downtime per year
99.99% uptime (four nines)	Up to 53 minutes of downtime per year
99.999% uptime (five nines)	Up to 5 minutes of downtime per year

Considering all the systems and network equipment commonly used in a business, it is simply not realistic to think that they can all run with five-nines availability. And when one considers that Windows servers require rebooting for such things as device driver updates, a person can only wonder how anybody can expect to have a five-nines operation.

The answer lies with redundancy: in doing more with more. The term "N+1" refers to a distributed system that requires N individual systems to do the work, but has a spare (+1) system that can take over if something happens to one of the other N systems.

So, by using extra systems, extra network links, and network equipment, as well as using extra storage connections and devices, it is possible to provide continuous application services, even when individual pieces of the puzzle break or are replaced.