Package org.h2.util

Class MemoryEstimator

java.lang.Object
org.h2.util.MemoryEstimator

public final class MemoryEstimator extends Object
Class MemoryEstimator. Calculation of the amount of memory occupied by keys, values and pages of the MVTable may become expensive operation for complex data types like Row. On the other hand, result of the calculation is used by page cache to limit it's size and determine when eviction is needed. Another usage is to trigger auto commit, based on amount of unsaved changes. In both cases reasonable (lets say ~30%) approximation would be good enough and will do the job. This class replaces exact calculation with an estimate based on a sliding window average of last 256 values. If estimation gets close to the exact value, then next N calculations are skipped and replaced with the estimate, where N depends on the estimation error.
  • Field Details

  • Constructor Details

    • MemoryEstimator

      private MemoryEstimator()
  • Method Details

    • estimateMemory

      public static <T> int estimateMemory(AtomicLong stats, DataType<T> dataType, T data)
      Estimates memory size of the data based on previous values.
      Type Parameters:
      T - type of the data
      Parameters:
      stats - AtomicLong holding statistical data about the estimated sequence
      dataType - used for calculation of the next sequence value, if necessary
      data - which size is to be calculated as the next sequence value, if necessary
      Returns:
      next estimated or calculated value of the sequence
    • estimateMemory

      public static <T> int estimateMemory(AtomicLong stats, DataType<T> dataType, T[] storage, int count)
      Estimates memory size of the data set based on previous values.
      Type Parameters:
      T - type of the data in the storage
      Parameters:
      stats - AtomicLong holding statistical data about the estimated sequence
      dataType - used for calculation of the next sequence value, if necessary
      storage - of the data set, which size is to be calculated
      count - number of data items in the storage
      Returns:
      next estimated or calculated size of the storage
    • samplingPct

      public static int samplingPct(AtomicLong stats)
      Calculates percentage of how many times actual calculation happened (vs. estimation)
      Parameters:
      stats - AtomicLong holding statistical data about the estimated sequence
      Returns:
      sampling percentage in range 0 - 100
    • calculateMagnitude

      private static int calculateMagnitude(long sum, long absDelta)
    • updateStatsData

      private static long updateStatsData(AtomicLong stats, long statsData, int counter, int skipSum, long initialized, long sum, int itemsCount, int itemsMem)
    • constructStatsData

      private static long constructStatsData(long sum, long initialized, int skipSum, int counter)
    • updateStatsData

      private static long updateStatsData(AtomicLong stats, long statsData, long updatedStatsData, int itemsCount, int itemsMem)
    • getCounter

      private static int getCounter(long statsData)
    • getSkipSum

      private static int getSkipSum(long statsData)
    • getAverage

      private static int getAverage(long updatedStatsData)