Performance: Fixes encoded strings performance

[fnarenji]

Pull Request Template

Description

This PR optimizes encoded string performance by:

• improve the algorithm that identifies the charset to remove redundant loops in TryEncodeCompressedString
• keep track of the offsets at which string reference indexes are stored to avoid a full-pass over all values of the document (& cuts allocations by 50%)

Before:

Method	Serializer	EnableEncodedStrings	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated
Write	Text	False	126.0 ms	2.44 ms	2.81 ms	1000.0000	-	-	75.67 MB
Write	Text	True	121.8 ms	2.28 ms	2.02 ms	1000.0000	-	-	75.67 MB
Write	BinaryNaive	False	241.4 ms	1.64 ms	1.37 ms	1000.0000	-	-	82.24 MB
Write	BinaryNaive	True	515.9 ms	5.00 ms	4.43 ms	3000.0000	-	-	183.48 MB
Write	BinaryBlitting	False	150.8 ms	3.01 ms	7.21 ms	1000.0000	-	-	71.71 MB
Write	BinaryBlitting	True	415.8 ms	3.27 ms	3.06 ms	3000.0000	-	-	172.95 MB

After:

Method	Serializer	EnableEncodedStrings	Mean	Error	StdDev	Median	Gen 0	Gen 1	Gen 2	Allocated
Write	Text	False	123.2 ms	1.96 ms	1.74 ms	122.7 ms	1000.0000	-	-	75.67 MB
Write	Text	True	121.3 ms	0.78 ms	0.65 ms	121.0 ms	1000.0000	-	-	75.67 MB
Write	BinaryNaive	False	249.7 ms	4.97 ms	8.30 ms	245.8 ms	1000.0000	-	-	86.15 MB
Write	BinaryNaive	True	278.5 ms	4.75 ms	4.45 ms	278.4 ms	1000.0000	-	-	99.1 MB
Write	BinaryBlitting	False	147.2 ms	2.89 ms	4.67 ms	146.7 ms	1000.0000	-	-	73.54 MB
Write	BinaryBlitting	True	178.1 ms	3.51 ms	5.14 ms	177.8 ms	1000.0000	-	-	86.49 MB

BinaryNaive exclusively uses public SDK primitives, whereas BinaryBlitting uses the typed JSON shortcuts. The EnableEncodedStrings has no impact on Text run.

Benchmark Updates

Text Navigator to Binary Writer

Before

Method	payload	sourceFormat	destinationFormat	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated
NavigatorToWriter	NutritionData	Text	Binary	34,552.14 us	619.581 us	579.557 us	1000.0000	333.3333	-	5572.44 KB

After

Method	payload	sourceFormat	destinationFormat	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated
NavigatorToWriter	NutritionData	Text	Binary	24,779.95 us	515.386 us	755.445 us	875.0000	500.0000	125.0000	5053.33 KB

30% reduction in time and 10% reduction in memory

Binary String Write MicroBenchmark

Before

Method	namedWriteDelegate	benchmarkSerializationFormat	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated
ExecuteWriteMicroBenchmark	string	Binary	1,023.25 ms	9.114 ms	8.525 ms	35000.0000	2000.0000	2000.0000	145.39 MB

After

Method	namedWriteDelegate	benchmarkSerializationFormat	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated
ExecuteWriteMicroBenchmark	string	Binary	452.70 ms	8.968 ms	15.229 ms	36000.0000	2000.0000	2000.0000	161.39 MB

half the time

Type of change

Closing issues

To automatically close an issue: closes #IssueNumber

[github-actions]

Please follow the required format: "[Internal] Category: (Adds|Fixes|Refactors) Description"

Examples:
Diagnostics: Adds GetElapsedClientLatency to CosmosDiagnostics
PartitionKey: Fixes null reference when using default(PartitionKey)
[v4] Client Encryption: Refactors code to external project
[Internal] Query: Adds code generator for CosmosNumbers for easy additions in the future.

All good!

[bchong95]

This code looks good to me. Once I get back to a computer I will run the benchmarks on my machine and update the baselines.

Thanks for making the change :)

[ealsur]

This is not Internal though, right? If it would benefit end users, so can we change the title?

[bchong95]

This is not Internal though, right? If it would benefit end users, so can we change the title?

No user uses binary writer, so it's still internal.

[bchong95]

Thanks for making the change. Very excited for the 2x speed up on the binary string write path!