Technology Evolution Insufficient
The evolution of the costs of compute, storage, and other key technologies becomes much less than anticipated, or is driven by markets rather than by technology.
We are unable to provide the needed level of resources with constrained budgets.
Ensure the core software of the experiments is as efficient as possible, also the experiments may need to prioritise their physics analysis programmes.
Ensure flexibility for both software and hardware choices. Software should optimize use of memory, CPU performance, I/O performance, etc. Computing models must optimize between hardware resources - CPU/storage/networks to minimize overall cost.
The evolution of "Moore's Law" over the past several decades, is no longer a good guide to the prediction of cost of resources. Several effects are in play:
- CPUs: while numbers of transistors still increases, they go into functions not easily usable by legacy HEP code;
- New types of processore are becoming more prevalent - GPU, etc. These do not follow the cost scaling models we have been used to.
- Many technologies (RAM, SSD, HDD, tapes - drives and media) are subject to markets (speculation, price holding) rather than technology driven. This leads to very unpredictable costs.
- New HDD technologies are very risky - may need to adapt the way they are used, they may need very different access models; in addition they could have very significant costs compared to todays HDD.
- We have seen recently that the canonical 20%/year price/performance improvement is very dependent on national conditions, often is no longer realisitic. For storage also depends on technology choices, and may be liited by national procurement boundary conditions.