Most supercomputers are designed for, and bought for, floating-point computations. Even here, it can be surprisingly difficult to get the advertised performance, as the author, a (super-)computer centre director observed. In particular, a great deal of performance depends on the regularity of the data, and exploiting the parallelism available within an individual core, between cores on anode, and between nodes. So how does the author, a computer algebraist, convince the director that he can make effective use of the supercomputer?