Reduce compression memory use with map[string]uint16 #852
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
map[string]uint16 uses 25% less memory per-entry than a map[string]int (16+2)/(16+8) = 0.75. All entries in the compression map are bound by maxCompressionOffset which is 14-bits and fits within a uint16.
Codecov Report
@@ Coverage Diff @@
## master #852 +/- ##
==========================================
+ Coverage 57.96% 58.05% +0.08%
==========================================
Files 42 42
Lines 10675 10690 +15
==========================================
+ Hits 6188 6206 +18
+ Misses 3398 3396 -2
+ Partials 1089 1088 -1
Continue to review full report at Codecov.
|
|
[ Quoting <notifications@github.com> in "[miekg/dns] Reduce compression memo..." ]
This PR switches the packing code to use a `map[string]uint16` for storing compression pointers while keeping support for external callers who pass in a `map[string]int`. This reduces the per-entry memory consumption of the compression map by 25%. (`(16+2)/(16+8) = 0.75`).
`compressionMap` is passed quite specifically as a value rather than a pointer because `RR.pack` is an interface method call which would cause it to escape and be heap allocated if it were a pointer. As maps are internally represented by a pointer to the runtime's `hmap` struct, the size of `compressionMap` is just 16-bytes anyway.
The benchmarks show a notable improvement in both memory use (alloc/op) and performance (time/op) for packing messages where multiple records are present.
```
name old time/op new time/op delta
MsgLengthPack-12 1.41µs ± 9% 1.46µs ±10% ~ (p=0.353 n=10+10)
PackDomainName-12 128ns ± 0% 137ns ± 1% +6.72% (p=0.000 n=6+10)
PackA-12 39.0ns ± 1% 40.1ns ± 1% +2.69% (p=0.000 n=8+8)
PackMX-12 69.6ns ± 1% 72.4ns ± 1% +4.05% (p=0.000 n=10+9)
PackAAAAA-12 37.9ns ± 1% 40.0ns ± 1% +5.68% (p=0.000 n=10+9)
PackMsg-12 1.01µs ± 5% 0.95µs ± 1% -5.85% (p=0.000 n=10+10)
PackMsgOnlyQuestion-12 168ns ± 0% 176ns ± 1% +4.94% (p=0.000 n=6+10)
PackMsgMassive-12 67.8µs ±10% 55.1µs ±10% -18.70% (p=0.000 n=10+10)
name old alloc/op new alloc/op delta
MsgLengthPack-12 576B ± 0% 528B ± 0% -8.33% (p=0.000 n=10+10)
PackDomainName-12 0.00B 0.00B ~ (all equal)
PackA-12 0.00B 0.00B ~ (all equal)
PackMX-12 0.00B 0.00B ~ (all equal)
PackAAAAA-12 0.00B 0.00B ~ (all equal)
PackMsg-12 256B ± 0% 208B ± 0% -18.75% (p=0.000 n=10+10)
PackMsgOnlyQuestion-12 0.00B 0.00B ~ (all equal)
PackMsgMassive-12 23.1kB ± 0% 19.0kB ± 0% -17.88% (p=0.000 n=10+10)
name old allocs/op new allocs/op delta
MsgLengthPack-12 3.00 ± 0% 3.00 ± 0% ~ (all equal)
PackDomainName-12 0.00 0.00 ~ (all equal)
PackA-12 0.00 0.00 ~ (all equal)
PackMX-12 0.00 0.00 ~ (all equal)
PackAAAAA-12 0.00 0.00 ~ (all equal)
PackMsg-12 2.00 ± 0% 2.00 ± 0% ~ (all equal)
PackMsgOnlyQuestion-12 0.00 0.00 ~ (all equal)
PackMsgMassive-12 12.0 ± 0% 12.0 ± 0% ~ (all equal)
```
Looking at these benchmarks are the skewed to larger messages? 'Cause most DNS
message are still small (less than 10 RRs)
|
|
Only If you look at the benchmarks, the increases are on the order of a few nanoseconds (and I believe come solely down to an extra few |
miekg
approved these changes
Dec 1, 2018
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.
This PR switches the packing code to use a
map[string]uint16for storing compression pointers while keeping support for external callers who pass in amap[string]int. This reduces the per-entry memory consumption of the compression map by 25%. ((16+2)/(16+8) = 0.75).compressionMapis passed quite specifically as a value rather than a pointer becauseRR.packis an interface method call which would cause it to escape and be heap allocated if it were to be a pointer. As maps are internally represented by a pointer to the runtime'shmapstruct, the size ofcompressionMapis only 16-bytes anyway.The benchmarks show a notable improvement in both memory use (alloc/op) and performance (time/op) for packing messages where multiple records are present.
Updates #652 (see #652 (comment)).
Edit: This is similar to #820 but for packing.