An Irish computer expert has an unlikely role in helping to fight crime, not in Ireland but in the USA. Dr John Murphy of the Electrical Engineering Department at Dublin City University is involved in a project to build the world's biggest fingerprint matching computer system.
The work is for the US Federal Bureau of Investigation which is investing $640 million (€605 million) on the system and which includes an array of 25 computers. The Integrated Automated Fingerprint Identification System (IAFIS) is a massive undertaking and will involve 3,000 staff. When completed, the centre in West Virginia will provide fingerprint and name identification services to law enforcement agencies across the US 24 hours a day, 365 days a year.
Great demands will be placed on the systems supporting this service, Dr Murphy said. The FBI's fingerprint database includes 40 million entries. IAFIS will be expected to be able to process a name search in just three seconds to support police at traffic stops or crime scenes. This does not sound so impressive until you realise that it will field a million of these requests per day on average.
There are expected to be tens of thousands of fingerprint match searches where the police believe they know the suspect. The system however is also expected to support "latent print" searches, full interrogations of the print database when there is no known suspect and a match is sought. Hundreds of these are expected to come in each day when IAFIS is fully up and running, Dr Murphy said. The IAFIS computer system will have to be extremely durable to handle such demands, Dr Murphy explained. It was thought too risky to install the full system and then decide whether it was up to the job, the typical way of developing a system. Instead the project organisers decided to develop a computer model that would mimic the operation of the real system.
This job fell to a Texas-based sub-contractor, Mayman International. Dr Murphy in turn became involved in the project through this company and won an 18-month research contract worth $50,000 to assist on preparation of the model.
The model, he said, doesn't search fingerprint files, it imitates the IAFIS computer system and its complex hardware and software architecture. The researchers can stress the model by simulating what would happen with heavy search volumes or storage operations and then watch for "cracks" to appear.
"What we are trying to do is find where these breakage points are," he said. "We have a dynamic computer model of this system. We go along and look at the general hardware architecture. We then look at the expected work load. The model isn't built and left. It is trying to anticipate where the problems will be."
If the model identifies a possible weak point in the system, this information is passed on to the system designers who can beef-up the structures and avoid breakdowns that might have caused the system to crash. "We aren't the designers, we are the performance assessors," Dr Murphy said.
The IAFIS designers and the FBI worry about what the system is expected to do. "That is not our concern. Ours is, can it do this fast enough and stay in one piece?
"The two main elements that I have been involved in is, when parts of IAFIS go down, what effect does that have on the overall system? I am also looking at how often this happens."
This approach to system preparation is known as "performance modelling" in the industry, he said. "This is mainly to do with probability distribution, queueing models and looking at computer code." It is a safer way to develop a large system and can prevent future problems. Parts of IAFIS are working and are used to check firearms licences. In this role it has already helped to solve a number of crimes, Dr Murphy said, although the full system isn't expected to go live until July this year.
The model develops as the system develops, he says, hence its value to the system designers. IAFIS has to be extremely flexible and will have peaks and troughs of activity through the working day. "It takes the whole night to recover and to be ready for the next day's transactions."
The fingerprint side of the system is based on an array of Convex computers, a company owned by Hewlett Packard. There are 25 of these computers with 16 processors in each machine. Purpose built "matching boards" have been developed and there will be 18 of these in each computer. All of this will be backed up by thousands of gigabytes of storage. The name searches are done using a Silicon Graphics computer with 28 processors. The modelling was particularly important in relation to this part of the project because of the three-second target response time. Britain already has a similar fingerprint matching system, Dr Murphy said. Its National Automated Fingerprint and Identification System was also performance modelled by Mayman International.
