Beware of drawing conclusions about generational GC from the Golang ISMM keynote. As far as I can tell, they didn’t test against copying generational GC with bump allocation in the nursery, which is the typical implementation.
-
-
-
Replying to @sgmansfield
How many non-copying generational GCs are actually deployed in the wild? I know of exactly one. All the others are copying.
1 reply 0 retweets 6 likes -
Replying to @pcwalton
"This is what everyone else is doing, therefore it's the right way?"
1 reply 0 retweets 0 likes -
Replying to @sgmansfield
Nobody has done benchmarks here, and in the absence of those I’m going to go with what’s been known for decades. Especially since there’s evidence that Go’s memory allocation is too slow for many people, since they manually rearrange code to satisfy escape analysis.
2 replies 0 retweets 10 likes -
Replying to @pcwalton
to satisfy your curiosity, you'd want an entirely new GC in the Go runtime? care to share the evidence you have?
1 reply 0 retweets 0 likes -
-
Replying to @sgmansfield1 reply 0 retweets 0 likes
-
Replying to @pcwalton
great! something to back up what you're saying. Now, what if instead they spent the effort on escape analysis instead of building a new GC in the runtime?
1 reply 0 retweets 0 likes
I suspect escape analysis has reached the point of diminishing returns. It’s easy to show that it needs higher-order control flow analysis to be precise—e.g. k-CFA, which is reallllly slow. (I did some work in this field several years ago.)
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.