Not sure if Kjell Winblad is around to answer questions, but I'm curious about the sharp performance drop-off (14M -> 5M) at 32 processes for the 50% write/ 50% delete benchmark. What's the bottleneck there?
Not directly related, but if I could have 2 thing from ETS, it would be:
1 - Ability to atomic rename a table. It's a fairly common use-case that you have a lock-free read-only map, rebuild it in the background, and the swap the newly-built one with the live one.
2 - Copy-on-write so that reading from an ETS doesn't require copying the data to the calling process (for more than just large binaries)
2 comments
[ 4.7 ms ] story [ 8.5 ms ] threadhttp://winsh.me/ets_catree_benchmark/azure_D64s_decent_ctrs/...
1 - Ability to atomic rename a table. It's a fairly common use-case that you have a lock-free read-only map, rebuild it in the background, and the swap the newly-built one with the live one.
2 - Copy-on-write so that reading from an ETS doesn't require copying the data to the calling process (for more than just large binaries)