While most issues with computers slowing down are well known, and pretty much a solved problem. There is one that is only slowly becoming noticeable as systems get progressively faster. This is the problem of memory slowing down as it ages. Every time a bit is flipped the underlying structures are stressed and an infinitesimal change takes place that makes it slightly more difficult for the next change. We are of course talking incredibly tiny amounts here so it's hardly surprising that it's not really been noticed up to now. However, it is something that has interested our most {cough} mature {cough} developer, Mary (she of ion trap fame). Once Mary gets interested in something there is simply no stopping her, so we just left her too it, and concentrated on our latest software build problem that only appears on the latest 'felt cowboy hat' distribution. Our Mary is a really methodical person, and she had this weird setup where she was rapidly switching RAM in various bit patterns and then every few hours, stopping, letting it rest then measuring the switching speed. This was fully under software control of course in an Automated Recursive Sequencing Environment. Once she had several sticks showing measurable speed drops (takes several months) she put half of them to one side as a control, then tried out every idea she could think of to get the others to 'relax'. The first was quite obviously thermal - the Applied Heat Neutralising Optimiser - but was inconclusive, and applied long enough it produced dead cells in the RAM. Then it was mechanical - her Vibrating Axis Guided Uniform Exciter idea. Nope, simply shook things apart. A sort of combination of both of these was to attack the problem with a Selective Acoustic Wavefront - quite ineffective unfortunately. Of course, some electronics can be altered with light, so Mary tried using Phased Optical Resonance Notching, although that raised a few eyebrows in the workshop. With no real breakthrough Mary moved on to trying software methods. Unfortunately I'm not permitted to explain how these work, so you'll have to try to work it out from their names. The first was Branched Asynchronous Selective Insert Code. Then there was the Binary Indexed Latency Extractor. The last of these was Consolidated Ram Accretion Pruner. Almost a year had passed now and Mary was showing visible signs of desperation (something quite shocking to witness). Almost in panic she tried combining hardware and software in just about every configuration. This morning she suddenly got up and almost ran out of the workshop without a word. She'd left her last scribbled note out though - it read: V.A.G.U.E Hardware OK, A.H.N.O C.R.A.P software tests A.R.S.E. We just hope she'll calm down by lunchtime. -- Will J Godfrey http://www.musically.me.uk Say you have a poem and I have a tune. Exchange them and we can both have a poem, a tune, and a song.