Skip to content

Home

Jump to bottom
Zeyuan Hu edited this page Apr 23, 2018 · 1 revision

The benchmarks page has a list of experiments evaluating PebblesDB vs LevelDB, HyperLevelDB, and RocksDB. The summary is that PebblesDB outperforms the other stores on write throughput, equals other stores on read throughput, and incurs a penalty for small range queries on fully compacted key-value stores. PebblesDB achieves 6x the write throughput of RocksDB, while providing similar read throughput, and performing 50% lesser IO. Please see the paper for more details.

If you would like to run MongoDB with PebblesDB as the storage engine, please check out mongo-pebbles, a modification of the mongo-rocks layer between RocksDB and MongoDB.

Running microbenchmark

  1. cd pebblesdb/
  2. make db_bench
  3. ./db_bench --benchmarks=<list-of-benchmarks> --num=<number-of-keys> --value_size=<size-of-value-in-bytes> --reads=<number-of-reads> --db=<database-directory-path>
    A complete set of parameters can be found in db/db_bench.cc

Sample usage:
./db_bench --benchmarks=fillrandom,readrandom --num=1000000 --value_size=1024 --reads=500000 --db=/tmp/pebblesdbtest-1000

Optimizations in PebblesDB

PebblesDB uses the FLSM data structure to logically arrange the sstables on disk. FLSM helps in achieving high write throughput by reducing write amplification. But in FLSM, each guard can contain multiple overlapping sstables. Hence a read or seek over the database requires examining one guard (multiple sstables) per level, thereby increasing the read/seek latency. PebblesDB employs some optimizations to tackle these challenges as follows:

Read optimization

  • PebblesDB makes use of sstable-level bloom filter instead of block level bloom filter used in HyperLevelDB or LevelDB. With this optimization, even though a guard can contain multiple sstables, PebblesDB effectively reads only one sstable from disk per level.

  • By default, this optimization is turned on, but this can be disabled by commenting the macro #define FILE_LEVEL_FILTER in db/version_set.h. Remember to do make db_bench after making a change.

Seek optimization

Sstable-level bloom filter can't be used to reduce the disk read for seek operation since seek has to examine all files within a guard even if a file doesn't contain the key. To tackle this challenge, PebblesDB does two optimizations:

  1. Parallel seeks: PebblesDB employs multiple threads to do seek() operation on multiple files within a guard. Note that this optimization might only be helpful when the size of the data set is much larger than the RAM size because otherwise the overhead of thread synchronization conceals the benefits obtained by using multiple threads. By default, this optimization is disabled. This can be enabled by uncommenting #define SEEK_PARALLEL in db/version_set.h.

  2. Forced compaction: When the workload is seek-heavy, PebblesDB can be configured to do a seek-based forced compaction which aims to reduce the number of files within a guard. This can lead to an increase in write IO, but this is a trade-off between write IO and seek throughput. By default, this optimization is enabled. This can be disabled by uncommenting #define DISABLE_SEEK_BASED_COMPACTION in db/version_set.h.

Tuning PebblesDB

  • The amount of overhead PebblesDB has for read/seek workloads as well as the amount of gain it has for write workloads depends on a single parameter: kMaxFilesPerGuardSentinel, which determines the maximum number of sstables that can be present within a single guard.

  • This parameter can be set in db/dbformat.h (default value: 2). Setting this parameter high will favor write throughput while setting it lower will favor read/seek throughputs.

Clone this wiki locally