hughk's Journal: Alphas and Unobtaniums

One of the organisations that I have worked for and with is a major electronic financial products exchange. Their host systems have been sitting on Alphas running VMS and have been since they transitioned off the VAX. As load has increased, they have normally just bough new hardware with only relatively limited architecture changes. Software changes carried risk, hardware changes within a hardware platform carried a minimal risk - so they just bought the latest model.

Unfortunately the Alpha is dead. The last model is the Alpha EV78 and then OpenVMS will be running on the Itanium, at least that seems to be what HP, the owners of Alpha and partners with Intel for the Itanium are promising.

When Digital introduced the Alpha, it took several years before it was seen to be reliable enough for use as a cluster server. It took the exchange even longer before they converted their application. Not because the application was hard to port, but more because the app was pushing some bits of VMS very hard and it certainly could find the weaknesses very easily.

The first Itaniums are out, but the word is that they are slow. Speed was never an issue with Alpha, but these babies start out with problems. A bad sign. VMS doesn't run properly on Itanium yet (it is past first boot, but that is all so the rumour goes), so I guess the conversion to Itanium is at least 5 years ahead. The app is currently very dependent on VMS but can't be stretched across large clusters due to the interaction between the individual cluster nodes. Scalability becomes an issue.

What to do about performance then, no new Alphas and a useful Itanium is a way away? Time to bite the bullet and do a rewrite?