Gibson Cache Server
In my work, I often have to use key-value storage for organizing communication between processes, storing system settings, temporary data, etc. For these purposes, I use Redis. He quite suits me, also hiredis to all libraries library.Literally today, I came across a completely new project - Gibson Cache Server . The first commit is dated May 17, 2013. What and speech!
Gibson Cache Server is an in-memory key-value database based on a data structure — a tree, while redis and memcached use a hash table. This allows you to use operations on multiple keys at once.
Features declared by the developer:
- Extremely fast
- Use as little memory as possible.
- Supports LZF compression
- Supports Time-To-Live (key lifetime)
- Supports key lock and unlock functionality
- Well, actually the most important thing is multiple operations.
Installation
Server
There is nothing complicated. Server sources can be found here .
git clone https://github.com/evilsocket/gibson.git cd gibson make -G «Unix Makefiles» . make sudo make install The configuration file is located in /etc/gibson/gibson.conf
By the way, the server can be started with the -c switch, indicating the path to the configuration file, and not installing everything in the system directories.
Customer
Sources are here .
Here, everything is the same. The only thing that should be noted in ./src/linenoise should be the source of the linenoise library.
After
git clone https://github.com/evilsocket/libgibsonclient cd libgibsonclient make -G «Unix Makefiles» . make We have a “gentleman's set”: the header file, the library itself and gibson-cli.
Demonstration
Run the client and the list of commands that are available at the moment.
Hidden text
deck@crunch ~/work/libgibsonclient $ ./gibson-cli -u /tmp/gibson.sock type :help or :h for a list of commands, :quit or :q to quit. local> :h SET <ttl> <key> <value> TLL <key> <ttl> GET <key> DEL <key> INC <key> DEC <key> LOCK <key> <seconds> UNLOCK <key> MSET <prefix> <value> MTTL <prefix> <ttl> MGET <prefix> MDEL <prefix> MINC <prefix> MDEC <prefix> MLOCK <prefix> <seconds> MUNLOCK <prefix> COUNT <prefix> STATS PING SIZEOF <key> MSIZEOF <prefix> ENCOF <key> Standard set / get
Hidden text
local> set 0 foo 5 <STRING> 5 local> get foo <STRING> 5 And using TTL.
Hidden text
local> set 3 bar hi! <STRING> hi! local> get bar <STRING> hi! local> get bar <REPL_ERR_NOT_FOUND> Well, multiple get / set
Hidden text
local> set 0 score:user1 400 <STRING> 400 local> set 0 score:user2 100 <STRING> 100 local> set 0 score:user3 900 <STRING> 900 local> mget score score:user1 => <STRING> 400 score:user2 => <STRING> 100 score:user3 => <STRING> 900 local> mset score 0 <NUMBER> 3 local> mget score score:user1 => <STRING> 0 score:user2 => <STRING> 0 score:user3 => <STRING> 0 ')
Performance
The developer’s site contains test data that he conducted on get / set operations on the machine.
Intel Core (TM) i3-2130 CPU @ 3.40GHz with 8GB of RAM, with Debian Squeeze.
He compared his database with Redis. The results are as follows:
Redis
Hidden text
$ redis-benchmark -c 1 -n 100000 -q -t SET SET: 51519.84 requests per second $ redis-benchmark -c 1 -n 100000 -q -t GET GET: 49212.60 requests per second Gibson
Hidden text
@ Created 100000 / 100000 in 1216ms -- 82236.842105 Req/s @ Verified : 100000 / 100000 in 1145ms -- 87336.244541 Req/s If you are not cunning, then the performance on these operations is 60 percent higher in Gibson.
Conclusion
Currently, there is already a library for php . Also, they promise to add Nodejs support soon.
The library can not be raw, but certainly deserves attention. Under my tasks should fit perfectly. Let's see what happens.
Source: https://habr.com/ru/post/185484/
All Articles