Reliability in computer systems

Reliability in computer systems

1. How is reliability measured?

   • Reliability can be measured: it has a 'metric'.
     • The metric tends to be an average measure of some kind.
       • AVAIL: this is the percentage of time that the system is available to the user, taking out any planned downtime (e.g. for maintenance).
         • For example, if in a planned uptime period of 1,000 hours the computer is available for 950 hours, its AVAIL figure is 95%.
       • MTTF: stands for Mean Time To Failure (also called MTBF, Mean Time Between Failure). This is the average number of hours something runs before it breaks, and is generally a measure for hardware such as a server.
         • For example if a hard disk fails on average after 500 hours then its MTTF figure is 500 hours.
       • There are other metrics of reliability, such as 'Probability of Failure of Demand'.

2. Ways of improving reliability

   • Redundancy: excellent method - duplicates critical parts of system so that if failure occurs the sub-system can take over. Very effective, but very expensive (basically building system twice).
     • Hardware redundancy: e.g. duplicating the power supply to a server - if the usual supply begins to fail due to component failure, the other power kicks in. Can be done for entire data centres - if mains power fails diesel generators kick in.
       • Personal computers can be protected with an uninterruptible power supply (UPS). This has a large battery that will keep the computer going for long enough to shut down in an orderly manner.
     • Data redundancy: duplicating the data in more than one place within the system. This is a good idea because if one component fails the data isn't lost.
       • Can be achieved by running two or more hard disks in parallel, with each storing the same data. This is called a 'RAID 1 array', or 'disk mirroring'. If one disk fails, the other keeps working and no data is lost. A technician will remove the broken disk and replace it - called 'hot swapping' because the server's power remains on.
     • Software redundancy: very rare, only used in safety critical applications.
       • Used because it is so difficult to create perfect code. Has three software routines, written by independent coding teams, that produce the same output given the same input. If there is no fault then all three modules run in parallel, but if one produces a different output then there is either an undiscovered bug in the software or the hardware it is running on has failed. In this case it is ignored as a control input and an error signal is sent.
         • Used for: medical equipment (e.g. patient monitoring), transport (anti-lock brakes, railway safety control), aviation (fly-by-wire systems), nuclear power station control software, and space (on-board satellite software).
   • Backing up: in order to continue from a failure it is essential that all the data is stored in a separate location. This is called 'data backup'.
     • Single machines: for personal, non-server computers, it is a good idea to have an attached external hard drive and a good backup application running on the machine.
     • Network server: for a file server, it is standard practice to back up to digital tape machines or to replicate the data on a remote server.
     • Data centre: for large organisations, an entire data centre may be duplicated in a separate location. This means that a fire or a flood in the main building will not result in data loss.
     • Cloud backup: one option is to hire an external specialist in data storage. The business/individual uploads their data every night to the remote data centre managed by the specialist. 'Cloud' computing and backup is becoming more popular as it is a cost effective solution, especially for small to medium sized companies (SMEs). However, there are growing concerns about the security of cloud storage.