Update CompressionBenchmark page sizes to realistic values (64K-1MB)

Use page sizes that reflect actual Parquet page sizes observed in practice:
64KB, 128KB, 256KB, and 1MB (the default). The 20K row-count limit
(PARQUET-1414) means most numeric columns produce pages of 78-234KB,
making the previous 8KB test point unrealistic.

Also fix JMH annotation processor path for Java 17+ compatibility
and reduce warmup/measurement iterations for faster iteration.

Performance results (master vs perf-compression-bypass branch):

Compression (ops/s, higher is better):
  Codec    | Page   | Master  | Branch  | Speedup
  SNAPPY   | 64KB   | 53,979  | 60,799  | +12.6%
  SNAPPY   | 128KB  | 27,764  | 30,524  |  +9.9%
  SNAPPY   | 256KB  | 13,549  | 14,648  |  +8.1%
  SNAPPY   | 1MB    |  2,445  |  2,675  |  +9.4%
  ZSTD     | 64KB   |  8,813  |  8,719  |  -1.1%
  ZSTD     | 128KB  |  4,361  |  4,501  |  +3.2%
  ZSTD     | 256KB  |  2,112  |  2,008  |  -4.9%
  ZSTD     | 1MB    |    423  |    422  |  -0.3%
  LZ4_RAW  | 64KB   | 37,777  | 36,107  |  -4.4%
  LZ4_RAW  | 128KB  | 16,777  | 16,330  |  -2.7%
  LZ4_RAW  | 256KB  |  9,060  |  8,956  |  -1.1%
  LZ4_RAW  | 1MB    |  1,961  |  2,191  | +11.7%
  GZIP     | 64KB   |  1,422  |  1,423  |  +0.1%
  GZIP     | 128KB  |    641  |    646  |  +0.8%
  GZIP     | 256KB  |    315  |    317  |  +0.7%
  GZIP     | 1MB    |     75  |     77  |  +2.3%

Decompression (ops/s, higher is better):
  Codec    | Page   | Master  | Branch  | Speedup
  SNAPPY   | 64KB   | 60,928  | 67,224  | +10.3%
  SNAPPY   | 128KB  | 29,919  | 33,457  | +11.8%
  SNAPPY   | 256KB  | 14,431  | 15,912  | +10.3%
  SNAPPY   | 1MB    |  3,140  |  3,540  | +12.7%
  ZSTD     | 64KB   | 32,042  | 35,750  | +11.6%
  ZSTD     | 128KB  | 19,447  | 21,800  | +12.1%
  ZSTD     | 256KB  |  9,495  | 10,759  | +13.3%
  ZSTD     | 1MB    |  2,155  |  2,409  | +11.8%
  LZ4_RAW  | 64KB   | 80,415  |118,358  | +47.2%
  LZ4_RAW  | 128KB  | 40,615  | 59,620  | +46.8%
  LZ4_RAW  | 256KB  | 19,888  | 29,914  | +50.4%
  LZ4_RAW  | 1MB    |  4,628  |  7,517  | +62.4%
  GZIP     | 64KB   |  9,393  |  9,608  |  +2.3%
  GZIP     | 128KB  |  4,101  |  4,536  | +10.6%
  GZIP     | 256KB  |  1,736  |  1,891  |  +8.9%
  GZIP     | 1MB    |    406  |    442  |  +9.1%

Key findings:
- SNAPPY: consistent 8-13% improvement across all page sizes
- LZ4_RAW decompression: strongest gain at 47-62% (eliminates 2x heap<->direct copies)
- ZSTD decompression: 11-13% from NoFinalizer + config caching
- GZIP decompression: 9-11% faster at 128KB+ page sizes
- ZSTD/GZIP compression: within noise (CPU-bound in native codec)
- LZ4_RAW compression: within noise at small pages, +12% at 1MB

Author: iemejia

parquet-benchmarks/pom.xml

@@ -89,6 +89,15 @@
       <plugin>
         <groupId>org.apache.maven.plugins</groupId>
         <artifactId>maven-compiler-plugin</artifactId>
+        <configuration>
+          <annotationProcessorPaths>
+            <path>
+              <groupId>org.openjdk.jmh</groupId>
+              <artifactId>jmh-generator-annprocess</artifactId>
+              <version>${jmh.version}</version>
+            </path>
+          </annotationProcessorPaths>
+        </configuration>
       </plugin>
       <plugin>
         <groupId>org.apache.maven.plugins</groupId>

parquet-benchmarks/src/main/java/org/apache/parquet/benchmarks/CompressionBenchmark.java

@@ -54,15 +54,15 @@
 @BenchmarkMode(Mode.Throughput)
 @OutputTimeUnit(TimeUnit.SECONDS)
 @Fork(1)
-@Warmup(iterations = 3, time = 2)
-@Measurement(iterations = 5, time = 3)
+@Warmup(iterations = 2, time = 1)
+@Measurement(iterations = 3, time = 2)
 @State(Scope.Thread)
 public class CompressionBenchmark {
 
   @Param({"SNAPPY", "ZSTD", "LZ4_RAW", "GZIP"})
   public String codec;
 
-  @Param({"8192", "65536", "262144"})
+  @Param({"65536", "131072", "262144", "1048576"})
   public int pageSize;
 
   private byte[] uncompressedData;