Understanding hash-max-zipmap-entries and "hash of hashes" optimization

When you set key in Redis, it creates some data structure, which involves pointer and some extra data.

Hashes are made in same way - data structure with ponters for each hash field / value.

However, in recent Redis versions (Redis 2.2), there is optimized way of storing hashes - it may be compared as PHP serialized string or like JSON string:

{"a":1,"b":2}

Note this is only example for better understanding, it does not represent internal way Redis uses.

When using this method, Redis may do huge savings of memory, however when access hash elements, it will need to iterate all hash field, e.g. Redis trades CPU for memory.

To control this behavour, Redis config have two parameters:

hash_max_zipmap_entries - if hash fields are less this value the optimized method is used.
hash_max_zipmap_value - if hash values sizes are less this value the optimized method is used.

There are similar config parameters not only for hashes, but also for sets and lists:

# conf parameter, default value
hash-max-zipmap-entries      64
hash-max-zipmap-value       512
list-max-ziplist-entries    512
list-max-ziplist-value       64
set-max-intset-entries      512

You can use those in redis.conf.

In Redis 2.4 there will be such setting for sorted lists as well.

Example use:

We made simple PHP code that utilize this optimization, along with some results.

The idea behind the code belong to Redis author - Salvatore Sanfilippo.

Instead of setting key=value, we calculate md5/sha1 from the key, get first 4 characters, and use it as hash name. In this hash we set field with original key name.

If we set 1M keys, there will be 65536 hashes, each with ~16 keys (1M / 65536). This is at least on theory.

Note if you decide to do so in production environments, you will not be able to do expiration and most of key functions must be performed in different way.

PHP code:

$r = new Redis();
$r->connect("127.0.0.1", "6379");

// Calculates the hash name...
function hkey($key){
	$maxlen = 4;
	$prefix = "_";

	return $prefix . substr(md5($key), 0, $maxlen);
}

// set key...
function hset($r, $key, $val){
	$key1 = hkey($key);

	return $r->hset($key1, $key, $val);
}

// get key...
function hget($r, $key, $val){
	$key1 = hkey($key);

	return $r->hget($key1, $key);

// ======================================

// Non optimized version
function hset1($r, $key, $val){
	return $r->set($key, $val);
}

for($i=0; $i < 1000000; $i++)
	hset($r, "key:" . $i, $i);}

The results:

Non optimized version first:

used_memory_human:57.92M
db0:keys=1000000,expires=0

Optimized version:

used_memory_human:22.40M
db0:keys=65536,expires=0

Here you are (default) config parameters we used for this test:

hash_max_zipmap_entries:512
hash_max_zipmap_value:64

Here you are a random element:

redis 127.0.0.1:6379> randomkey
"_d488"
redis 127.0.0.1:6379> hgetall  "_d488"
 1) "key:50348"
 2) "50348"
 3) "key:101947"
 4) "101947"
 5) "key:128127"
 6) "128127"
 7) "key:260902"
 8) "260902"
 9) "key:292977"
10) "292977"
11) "key:338186"
12) "338186"
13) "key:359584"
14) "359584"
15) "key:441871"
16) "441871"
17) "key:619595"
18) "619595"
19) "key:630408"
20) "630408"
21) "key:876941"
22) "876941"
23) "key:918003"
24) "918003"
25) "key:931704"
26) "931704"

Links:

blog comments powered by Disqus

Understanding hash-max-zipmap-entries and "hash of hashes" optimization

Example use:

PHP code:

The results:

Links:

Articles library

Redis as session handler in PHP

Lazy starter script on Linux such Redhat CentOS or Fedora

Redis vs Memcached comparison

Why RDBMS and SQL are difficult...

Seamless migration from one Redis server to another

Redis swap issue while save

Amazon EC2 and Amazon ElastiCache service

Understanding hash-max-zipmap-entries and 'hash of hashes' optimization

NoSQL database design example

Redis save and backup script

Postgres 9.1 foreign data wrapper interface

Redis high traffic connection issue

Redis too many open files error on high traffic sites

PHP Redis bug with PHP 5.4

Case Study: Using Redis intersect on very large sets

Getting started with Python and Redis on CentOS 6.X

GeoIP in Redis

Redis database file examination