YoDB

A lightweight and fast key-value storage engine based on the buffer tree.

Purpose

There have already been several KV databases like tokudb and cascadb, which are using buffer tree as the underlying data structure to optimize written operation. But their code seems to be a little complex and hard for the beginners to understand the core idea of how the tree really works in multi-threaded environment.

So I write this storage engine which named yodb (yo just helps for funny pronunciation), try to meet the need of those guys who want to have a quick introspect of this beautiful algorithm. Yodb has an excellent performance that can handle millions of read/written requests at a time with only 6K source lines of code, and also of course has a detailed notation.

Performance

Setup

We use a database with a million entries. Each entry has a 16 byte key, and a 100 byte value.

yodb:       version 0.1
Date:       Tue Dec 17 15:00:09 2013
CPU:        4 * Intel(R) Core(TM)2 Quad CPU    Q8300  @ 2.50GHz
CPUCache:   2048 KB
Keys:       16 bytes each
Values:     100 bytes each 
Entries:    1000000
RawSize:    110.6 MB (estimated)
FileSize:   110.6 MB (estimated, compression disabled)

Write performance

fillseq      :       4.989 micros/op;   22.2 MB/s     
fillrandom   :       5.223 micros/op;   21.2 MB/s

Each "op" above corresponds to a read/write of a single key/value pair. I.e., a random write benchmark goes at approximately 200,000 writes per second.

Read performance

readseq      :       2.653 micros/op;   41.7 MB/s  
readrandom   :       7.804 micros/op;    6.4 MB/s  
readhot      :       2.662 micros/op;   41.6 MB/s

Usage

Include to your header

#include <yodb/db.h>

using namespace yodb;

Options opts;
opts.comparator = new BytewiseComparator();
opts.env = new Env("/your/database/path");

DB* db = new DB("your_db_name", opts);
if (!db->init()) {
    fprintf(stderr, "error initialize database\n");
}

Write

if (!db->put("Shanghai", "Minhang part")) {
    fprintf(stderr, "insert error\n");
}

Read

Slice value;
if (!db->get("Guangzhou", value)) {
    fprintf(stderr, "read error\n");
}

Delete

if (!db->del("Beijing")) {
    fprintf(stderr, "delete error\n");
}

Exit

delete db;
delete opts.comparator;
delete opts.env;

Further work

Make lock independent with the tree.
Add bloom filter to accelerate read operation.
Add memory table just like leveldb

kedebug/yodb