Move to String-backed HPACK representations. #89
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Motivation: When originally written our HPACK implementation targetted Swift 4.2, which had performance issues when storing mostly-ASCII data in Strings. For this reason we had a number of data types that used ByteBuffers as their backing storage and reified Strings when necessary. In Swift 5 this is no longer worthwhile: in fact, it imposes performance penalties above and beyond simply using the Strings. For this reason, we should move to a simpler model where we simply store Strings again. Modifications: - Rewrote all HPACK data structures to use Strings as their backing store. - Removed now-redundant genericism. - Removed now-unnecessary StringRing data structure and associated tests. Result: Faster performance in almost all cases.
|
This change provides a substantial performance boost. I benchmarked using h2load to provide a fairly unrepresentative workload (doing a bunch of GETs and sending back almost no body) which puts a lot of strain on the HPACK implementation. Before this patch: After this patch: That's a more than 2x speedup, as well as a substantial decrease in latency in any case that was using indexed headers. |
|
AMAZING |
|
Fantastic work. All that messing about with byte containers was one of the biggest headaches for me I putting it together (though in all fairness I did learn a lot in the process). Very nice to be able to have a sane String-based implementation now.
…Sent from my iPhone
On Apr 10, 2019, at 9:10 AM, Cory Benfield ***@***.***> wrote:
This change provides a substantial performance boost. I benchmarked using h2load to provide a fairly unrepresentative workload (doing a bunch of GETs and sending back almost no body) which puts a lot of strain on the HPACK implementation. Before this patch:
min max mean sd +/- sd
time for request: 359.58ms 508.06ms 400.98ms 23.49ms 62.69%
req/s : 229.99 260.83 248.95 13.88 64.00%
After this patch:
min max mean sd +/- sd
time for request: 116.21ms 428.62ms 187.57ms 33.18ms 86.33%
req/s : 481.62 565.77 532.60 31.92 57.00%
That's a more than 2x speedup, as well as a substantial decrease in latency in any case that was using indexed headers.
—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub, or mute the thread.
|
|
Yup, it was very subtle code and being able to performantly remove it is a big win. It also reduces the surface area for further optimisation, so profiling should do a better job finding expensive operations. |
Lukasa
added
semver/minor
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Motivation:
When originally written our HPACK implementation targetted Swift 4.2,
which had performance issues when storing mostly-ASCII data in Strings.
For this reason we had a number of data types that used ByteBuffers as
their backing storage and reified Strings when necessary.
In Swift 5 this is no longer worthwhile: in fact, it imposes performance
penalties above and beyond simply using the Strings. For this reason, we
should move to a simpler model where we simply store Strings again.
Modifications:
Result:
Faster performance in almost all cases.