Just have a quick question -- is $B\Delta I$ compression actually used in the industry, or is it just a prototype in academia? This seems like a really viable compression scheme, and should also be easy to implement (I'm no ECE major, so it's just a guess).
Would this scheme increase the complexity of the caching model? If all lines are of varying length, determining the usable cache size is difficult for programmers. This might adversely impact performance day to day. For example, on one day our pointers get mapped to a virtual address space which has a low address and on another day they are mapped to a higher address, causing worse compression and giving vastly different performance numbers if the working set grows beyond the cache size because of worse compression.
patrickbot
The added complexity is math operations, which are cheap. The performance should be in the worst case still comparable with uncompressed performance, so I think compressing is better than not compressing.
Just have a quick question -- is $B\Delta I$ compression actually used in the industry, or is it just a prototype in academia? This seems like a really viable compression scheme, and should also be easy to implement (I'm no ECE major, so it's just a guess).
It is academic research.
http://www.cs.cmu.edu/~gpekhime/Papers/pact12_pekhimenko.pdf
Would this scheme increase the complexity of the caching model? If all lines are of varying length, determining the usable cache size is difficult for programmers. This might adversely impact performance day to day. For example, on one day our pointers get mapped to a virtual address space which has a low address and on another day they are mapped to a higher address, causing worse compression and giving vastly different performance numbers if the working set grows beyond the cache size because of worse compression.
The added complexity is math operations, which are cheap. The performance should be in the worst case still comparable with uncompressed performance, so I think compressing is better than not compressing.