TAJO-1716: Repartitioner.makeEvenDistributedFetchImpl() does not distribute fetches evenly

[jihoonson]

Here is the evaluation result.

• of samples: 1000

  • The sample size follows the zipf distribution in the range [1, 100].

• of fetches: 2000

Before patch

# of partitions: 10
std dev: 7239.203108077597
# of partitions: 20
std dev: 2507.946123823235
# of partitions: 30
std dev: 1337.669359578645
# of partitions: 40
std dev: 847.1299428068963
# of partitions: 50
std dev: 590.6100520648041
# of partitions: 60
std dev: 441.75456483838553
# of partitions: 70
std dev: 340.5741442844647
# of partitions: 80
std dev: 270.92110936581224
# of partitions: 90
std dev: 221.5819175222533
# of partitions: 100
std dev: 184.4234215060522

After patch

# of partitions: 10
std dev: 13.61763562654033
# of partitions: 20
std dev: 21.6693331706813
# of partitions: 30
std dev: 15.134361197689614
# of partitions: 40
std dev: 13.04185569631777
# of partitions: 50
std dev: 14.063911262269487
# of partitions: 60
std dev: 12.51514637879624
# of partitions: 70
std dev: 12.309312883565598
# of partitions: 80
std dev: 29.554991118293103
# of partitions: 90
std dev: 37.39714123885218
# of partitions: 100
std dev: 11.820253804288662

[jinossy]

Big +1!!

[jihoonson]

Thanks for your review!