No ... I m not falling in love with Microsoft
but I have to admit they have interesting open source projects :-) True they have great interest supporting FreeBSD, that's a given ; but that is a great benefit regardless I think ... While mimalloc has a well deserved "summer time", I started to look somewhere else in the meantime and stumbled upon snmalloc, which sounded promising especially in multithread context, just looking at the description.
Benchmarking this is always very tricky, always depends on the nature of the software you override memory allocation with, you can always come up with a conclusion while your close neighbour would shake your shoulder and showing you a very contradictory output in his side ...
Nevertheless, I gave a try with several professional and personal projects of mine and what I notice usually is ... snmalloc performs a bit better than hoard but below mimalloc (still remarkable from the rest of the "pack"), with single thread applications. However, snmalloc is above the latter in multi thread context. More generally, my tests displayed they all perform better than the Linux's system allocator in the vast majority of the case. I usually played with hoard but the development is sort of frozen since a while (diehard from the same author is a nice toy to study by the way) or jemalloc before ; but those new players give nice extra choices to look at. As a result of my appreciation, I just pushed a basic openbsd support proposal :-)
This week had been a nice surprise to see American Fuzzy Lop "resurrected" on github in the official google account. Indeed, there was not much activities since the well known 2.52b version. For sure, libFuzzer from LLVM is great but like always, it is better to have multiple screwdrivers around you and I kinda like the sense of humour of this software, if you look carefully while it runs you will understand :-) but the results are very serious, its reputation is well deserved. Now trying to bring improvements ideas for BSD and mac I had since long months ; we will see.
Apart of this, LLVM 9, the first Apache2 licensed version, should occur end of August ... September maybe if more bugs are found. One noticeable thing in FreeBSD's side, for the future LLVM 10, there is a possible interesting fix ongoing from a famous CHERIBSD contributor, hopefully will make it soon-ish and I think it really deserves to be backported to the 9's line. Redis had merged my little FreeBSD support improvement, nodejs as well accepted my tiny OpenBSD build fix.
As American Fuzzy Lop would say "We're done here. Have a nice day !"
Labels: FreeBSD, hoard, jemalloc, llvm, mimalloc, nodejs, OpenBSD, snmalloc