Photonics and future datacenter networks
Al Davis · 2010
- Advances in electronics will continue BUT • processing benefits from these advances • data center communications will benefit but not as much • optics is the transport choice, electronics is the processor choice in an ideal world - NOTE: we don't live in an ideal world - Complete change to optical communication will not happen in one step • e.g. multi-core was a tough bridge for merchant semiconductors to cross - argument with Albert Yu in 2000 but Kunle had presented the case well in 1996 - Tejas cancelled in 2004 - note the 8 year lag between research and industry adoption • industry momentum is significant but so is the research side - Power wall is here to stay (I don't see the magic technology which moves the wall) • going green is not going to be easy if consumption is based on MORE • getting more performance for less power is problematic • replacing long wires with optical paths is a good idea - telecomm did this in the 80's - definition of long for computing is changing however • maybe it should be relative to transistor speed