3 ms·
2% better perf isn't anything to sneeze at but also not super drastic. Will be curious to see if this unlocks more optimizations over time. I'm also curious if
by vlovich123 6y ago
2% better perf isn't anything to sneeze at but also not super drastic. Will be curious to see if this unlocks more optimizations over time.
I'm also curious if there's been any work to organize code layout to make it more efficient to page stuff out efficiently to reduce memory pressure (so that you're using all code in the pages you are bringing in) & reduce paging thrash (you're not evicting a code page only to bring it back for a few hundred bytes).
- hinkley 6y agoGiven the greatly improved TLB hit rate, I sort of wonder if the test isn't abusing the code enough. Does this change the inflection point where throughput falls from trying to run too many tasks in parallel or sequentially? 2% response time won't do much, but 10% higher peak load would change quite a few priorities on a lot of backlogs.
- vlovich123 6y agoIt’s not too surprising to me. If TLB misses cost your program to run 6% slower, wouldn’t you expect a 30% reduction in that to be a 2% overall improvement? Why would it be more? If you’re surprised that TLB misses are not so costly, consider that cache locality and CPU prediction are largely going to hide the cost of misses (cache locality means the miss rate is low to begin with, prediction reduces the cost of a miss). TLBs are also super complex involving a few layers of caching (TLB look aside, L1 cache of your TLB table, L2 cache, L3 and then RAM).