Breaking the Availability Barrier
Survivable Systems for Enterprise Computing
by
Book Cover & Preview Text
Chapter 1
The 9s Game
True or False:
- Adding processors to a multi-processor redundant system increases
its reliability.
- A 16-processor redundant system has the reliability of a UNIX box.
- A fault-tolerant processor board is less reliable than a UNIX processor
board.
The above are indeed provocative questions, and
their answers are not straightforward. This chapter develops some simple,
though not necessarily intuitive, concepts that help answer these and other
questions. More importantly, such concepts lead to some straightforward steps
that you can take to improve the reliability of your system – steps as simple
as being aware of how you allocate processes to processors.
The questions above all refer to reliability.
But before we go much further, we have to agree on how to measure reliability.
There are actually two components that impact the
reliability of a system – how long it will work before it fails and then how
long it will take to fix. We call the first component the mean time before
failure (MTBF) and the second the mean time to repair (MTR).
To a space satellite designer, reliability is MTBF.
Once the satellite fails, it is gone forever. It is not repairable. Clearly, a
satellite with a ten-year MTBF is ten times more reliable than a satellite with
a one-year MTBF.
However, when it comes to life and property
protection, reliability is MTR. In a 911 system, an outage of 30 seconds may be
simply an aggravating hiatus; but an outage of one hour can mean death by
cardiac arrest or a building burned to the ground.
In large transaction processing systems, reliability
is often measured as down time. Down time has a cost associated with it –
perhaps $1,000 per hour or $100,000 per hour. (Of course, MTR plays a role here
as well; the longer the down time, the higher the cost in many cases – from
customer annoyance to lost sales to lost customers.) Down time alternatively
can be measured as the proportion of time that a system is up, a measure that
we call the availability of the system.
Formats
Book Details
About the Book
As our daily lives and corporate well-being become more dependent upon computers, system reliability grows increasingly important. No longer are frequent system outages acceptable. In many cases, failure intervals must now be measured in centuries.
Even current fault-tolerant computing systems will fail once every five or ten years. This book is the first in a three-part series on active/active systems. It describes techniques that can be used today for extending system failure times from years to centuries, often at little or no additional cost.
The techniques described include splitting a large system into smaller, cooperating independent nodes. Copies of the application’s database are distributed across the nodes. It is shown that these techniques significantly reduce the number of system failure modes and increase the level of sparing. As a result, the loss of a single node’s capacity occurs far less frequently than the loss of all capacity when the equivalent monolithic system fails. Furthermore, the loss of more than one node’s worth of capacity is almost never.
Central to these techniques is the requirement that all database copies that are distributed across the network must be kept in synchronism. Several methods available today for maintaining synchronism are described. They include asynchronous data replication, synchronous data replication, and network transactions.
About the Author
Dr. Bill Highleyman, Paul J. Holenstein, and Dr. Bruce Holenstein have a combined experience of over 90 years in the implementation of fault-tolerant, highly available computing systems. This experience ranges from the early days of custom redundant systems to today’s fault-tolerant offerings from HP (NonStop) and Stratus.
Dr. Bill Highleyman has done extensive work on the effect of failure mode reduction on system availability. He has built fault-tolerant systems for train control, racetrack wagering, securities trading, message communication, and other applications. He is the Managing Editor of the Availability Digest (availabilitydigest.com).
Paul J. Holenstein and Dr. Bruce Holenstein have architected and implemented the various data replication techniques required for the availability enhancements described in this book. Their company, Gravic, provides the Shadowbase line of data replication products to the fault-tolerant community.