This benchmark measures the performance of RocksDB single key
read and write operations that are interleaved.
The keys are InodeIDs and the values are the Inode themselves.
Each thread creates mFileCount key/value pairs, then performs
operations over these, adds and deletes will be performed at
approximately the same rate to keep the total number of files
The following options can be varied.
mWritePercentage - the percentage of write operations, half of these will
be adds and half will be deletes, the remaining will be reads.
mWriteSerialization - enable or disable ProtoBuf serialization of writes.
mUserSerialization - enable or disable ProtoBuf serialization of reads.
mFileCount - the initial number of inodes stored in RocksDB for each
mIsDirectory - have the Inodes represent directories or files.
mUseZipf - if to use a Zipfian distribution when choosing the keys to read.
The more likely keys to be read will be the ones written last,
meaning that they will more likely be in the RocksDB memtable.
mRocksConfig - see RocksBenchConfig