-
-
Notifications
You must be signed in to change notification settings - Fork 2
Conversation
Codecov ReportAttention:
Additional details and impacted files@@ Coverage Diff @@
## main #6 +/- ##
==========================================
- Coverage 77.77% 71.42% -6.35%
==========================================
Files 1 1
Lines 18 21 +3
Branches 4 4
==========================================
+ Hits 14 15 +1
- Misses 2 4 +2
Partials 2 2 ☔ View full report in Codecov by Sentry. |
As a general comment: isa-l compresses to much lower compression levels by default. In case that is undesirable, it is better to use zlib-ng. On the other hand, if deflate compression is bottlenecking, shaving another 20% of the filesize is not really worth the delay in latency sensitive situations. The difference in speed between zlib-ng level 2 and isa-l level 2 will be much less than the difference between isa-l level 2 (default) and zlib-ng level 6 (default). zlib-ng level 1 is much bigger than either zlib level 1 or isa-l level 1, at least on genomic data. So I don't recommend using that. |
Seems to be a better trade off with using isal. It's faster and we get better compression for the websocket use case |
Then they are probably already using level 1. Then ISA-L is so much better. The compression speed is incredible, almost as fast as the decompression speed, 5 times faster than zlib. It really is an incredible library. I prefer using ISA-L where I can, and I maintain python-zlib-ng only because python-isal cannot cover all use cases. So all the cool innovations happen in python-isal first and are then ported over later when they proved stable. |
I wonder, have you considered porting this work over to aiohttp directly? In that way everyone will benefit from your work. |
That's the eventual plan. Usually I try out changes in libraries for a few months before upstreaming them into aiohttp. |
isal recommended for speed by pycompression authors pycompression/python-zlib-ng#24 (comment)
Tested on x86_64, macos arm64, and aarch64