CascaDB

Yet another write-optimized storage engine, using buffered B-tree algorithm inspired by TokuDB.

Features

• Provides a key-value access API similar to LevelDB.
• Support Snappy compression.
• Support Direct IO and Linux AIO.

Dependencies

CascaDB can have better performance if libaio and google snappy library 're installed. Otherwise Posix AIO (simulate AIO with multiple threads which is not true asynchronous) is used instead and data blocks're not compressed.

• libaio

  On Ubuntu Linux

    sudo apt-get install libaio-dev
  

  On RHEL

    sudo yum install libaio-devel
  

• snappy

Compile

CascaDB utilizes CMake to build, so first of all you should install CMake.

It's recommended to make out of source build, that is, object files are generated into a separated directory with the source directory.

    mkdir build
    cd build
    cmake ..
    make && make install

Performance

Running bin/db_bench to get detailed performance report of how CascaDB running on your machine.

e.g. bin/db_bench --num=1000000, This will create and test on a database of a million records, each record has a 16 byte key, and a 100 byte value.

Write Performance

    fillseq      :       1.020 micros/op;                 
    fillrandom   :       2.559 micros/op;                 

Read Performance

Read benchmark results after sequential insertions

    readrandom   :       1.125 micros/op;                 
    readseq      :       1.282 micros/op;      

Read benchmark results after random insertions

    readrandom   :       3.227 micros/op;                 
    readseq      :       1.295 micros/op;                 

Have a try!

1. Include the header file.

    #include <cascadb/db.h>
   

2. Open the database.

    Options opts;
    opts.dir = create_fs_directory("/tmp/db_bench");
    opts.comparator = new LexicalComparator();
    opts.compress = kSnappyCompress;
   
    const char *dbname = "example";
    DB *db = DB::open(dbname, opts);
    if (!db) {
      fprintf(stderr, "open error %s\n", dbname);
      exit(1);
    }
   

3. Insert record.

    if (!db->put("This is the key", "This is the value") {
        fprintf(stderr, "put error\n");
    }
   

4. Read record.

    Slice value;
    if (!db->get("This is the key", value) {
        fprintf(stderr, "get error\n");
    }
    fprintf(stdout, "the value is %s", value.to_string().c_str());
   

5. Delete record.

    if (!db->del("This is the key")) {
        fprintf(stderr, "del error\n");
    }
   

6. Destory database.

    delete db;
    delete opts.comparator;
    delete opts.dir;
   

TODO

• Add forward and backward iteration over data.
• Support more compression methods.
• Implement write ahead log to ensure write operation is atomic.