Why we still need MyISAM (for read-only tables)

TL;DR: we still need MyISAM and myisampack because it uses less space on disk (half of compressed InnoDB) !

In the previous post, I shared my experience with InnoDB table compression on a read-only dataset.  In it, I claimed, without giving much detail, that using MyISAM and myisampack would result is a more compact storage on disk.  In this post, I will share more details about this claim.

MyISAM is the original storage engine of MySQL.  It is a modified version of ISAM for MySQL (My-ISAM).  It was replaced by InnoDB as the default storage engine in MySQL 5.5 (first GA release in December 2010).  MyISAM is sometimes qualified as "not under active development", unsupported, unmaintained, deprecated, end-of life, and/or other unflattering adjectives.  There are even people writing about the end of MyISAM.  However in some situations and even if it does not support good concurrency, neither transactions nor foreign keys, MyISAM (and myisampack) can still be very useful.  The end of MyISAM would leave a gap in MySQL feature-set.  Let's see why.

In the previous post, I described a replicated system storing thousands of read-only tables.  Those tables are written as non-compressed InnoDB table on the master and they reach slaves via standard replication.  After being written, those tables are converted locally (not writing to the binary logs) to compressed InnoDB tables.  This post-compression significantly reduces the storage needed for this system.

MyISAM and myisampack could replace InnoDB table compression on this system.  The tables would still be written on the master using the InnoDB storage engine (for replication crash safety), but MyISAM and myisampack would be used instead of InnoDB compression for reducing disk footprint.  As compression is done on a copy of the table (atomically swapping/renaming tables when completed), the lack of transaction support in MyISAM is not a problem: if a crash happens during compression, the operation can be restarted from the beginning without losing any data.  Also, as the tables are read-only, compressed MyISAM tables can be used (after packing a table using myisampack, the table becomes read-only).

We can use myisampack because our dataset is read-only !

In the previous post, I described a log table where I am keeping information about the size of each compression round.  For testing MyISAM and myisampack, I used an extra slave where I added four fields to the compression log table: myisam_size, myisam_size2, mypack_size and mypack_size2.  The structure of table becomes the following:

> show create table compression_logs\G
*************************** 1. row ***************************
       Table: compression_logs
Create Table: CREATE TABLE `compression_logs` (
  `id` int(10) unsigned NOT NULL AUTO_INCREMENT,
  `schema_name` varchar(64) NOT NULL,
  `table_name` varchar(64) NOT NULL,
  `initial_size` bigint(20) unsigned NOT NULL,
  `state` enum([...]) NOT NULL DEFAULT 'copying',
  `copied_size` bigint(20) unsigned DEFAULT NULL,
  `kbs_8_size` bigint(20) unsigned DEFAULT NULL,
  `kbs_4_size` bigint(20) unsigned DEFAULT NULL,
  `kbs_2_size` bigint(20) unsigned DEFAULT NULL,
  `kbs_1_size` bigint(20) unsigned DEFAULT NULL,
  `final_size` bigint(20) unsigned DEFAULT NULL,
  `final_kbs` tinyint(4) DEFAULT NULL,
  `myisam_size` bigint(20) unsigned DEFAULT NULL,
  `myisam_size2` bigint(20) unsigned DEFAULT NULL,
  `mypack_size` bigint(20) unsigned DEFAULT NULL,
  `mypack_size2` bigint(20) unsigned DEFAULT NULL,
  PRIMARY KEY (`id`)
) ENGINE=InnoDB
1 row in set (0.01 sec)

We need two fields for MyISAM and myisampack sizes because, contrarily to InnoDB that stores data in a single .ibd file, MyISAM and myisampack store data in two files: .MYD for the data and .MYI for indexes.  I could have stored the sum of both values in a single field but this would have lost some information that could be useful in the future (feel free to ask me creative questions/queries on that data in the comments).

Before going further, let's look again at the InnoDB statistics for the MySQL instance analysed in the previous post:

> SELECT nb_tables, initial_gib, final_gib,
    ->   ROUND(initial_gib / final_gib, 2) AS ratio
    -> FROM (SELECT COUNT(*) AS nb_tables,
    ->              ROUND(SUM(initial_size) / POW(2,30), 2) AS initial_gib,
    ->              ROUND(SUM(final_size)   / POW(2,30), 2) AS final_gib
    ->       FROM compression_logs) AS c;
+-----------+-------------+-----------+-------+
| nb_tables | initial_gib | final_gib | ratio |
+-----------+-------------+-----------+-------+
|    248670 |    20701.84 |   4566.50 |  4.53 |
+-----------+-------------+-----------+-------+
1 row in set (0.15 sec)

This MySQL instance:

• contains 248,670 tables
• which would use 20.21 TiB of disk space with uncompressed
• but they are only using 4.45 TiB after compression
• so InnoDB table compression divides storage requirements by 4.53.

How would MyISAM and myisampack compare to above ?  To answer that question, I ran a script that, for each table:

• takes a MyISAM copy of the table,
• saves the MyISAM table sizes in the log table (.MYD and .MYI),
• runs myisampack and myisamchk on the table,
• and saves the myisampack-ed sizes in the log table.

As shown by the query below and in average, InnoDB uncompressed is using 22% more disk space than MyISAM:

> SELECT nb_tables, initial_gib, myisam_gib,
    ->   ROUND(initial_gib / myisam_gib, 2) AS ratio
    -> FROM (SELECT COUNT(*) AS nb_tables,
    ->              ROUND(SUM(initial_size)             / POW(2,30), 2) AS initial_gib,
    ->              ROUND(SUM(myisam_size+myisam_size2) / POW(2,30), 2) AS myisam_gib
    ->       FROM compression_logs) AS c;
+-----------+-------------+------------+-------+
| nb_tables | initial_gib | myisam_gib | ratio |
+-----------+-------------+------------+-------+
|    248670 |    20701.84 |   16906.99 |  1.22 |
+-----------+-------------+------------+-------+
1 row in set (0.14 sec)

As we have an average compression ratio of 4.53 on this dataset with InnoDB table compression, MyISAM alone is not very interesting.  The query below shows more interesting results with myisampack:

> SELECT nb_tables, initial_gib, final_gib, mypack_gib,
    ->   ROUND(initial_gib / mypack_gib, 2) AS r_mypack,
    ->   ROUND(final_gib   / mypack_gib, 2) AS r_i_vs_m
    -> FROM (SELECT COUNT(*) AS nb_tables,
    ->              ROUND(SUM(initial_size)             / POW(2,30), 2) AS initial_gib,
    ->              ROUND(SUM(final_size)               / POW(2,30), 2) AS final_gib,
    ->              ROUND(SUM(mypack_size+mypack_size2) / POW(2,30), 2) AS mypack_gib
    ->       FROM compression_logs) AS c;
+-----------+-------------+-----------+------------+----------+----------+
| nb_tables | initial_gib | final_gib | mypack_gib | r_mypack | r_i_vs_m |
+-----------+-------------+-----------+------------+----------+----------+
|    248670 |    20701.84 |   4566.50 |    1987.52 |    10.42 |     2.30 |
+-----------+-------------+-----------+------------+----------+----------+
1 row in set (0.17 sec)

So myisampack average compression ratio on the initial size (InnoDB uncompressed) is 10.42, which is 2.3 times better than InnoDB compressed.  I do not have formal data on the next statement, but I also think that myisampack takes less time to compress a table, especially that with InnoDB and to obtain the best compression ratio, we have to try many KEY_BLOCK_SIZE, which takes time.

One last thing before closing: let's check if there are cases where MyISAM or myisampack is using more disk space than InnoDB and InnoDB compression:

> SELECT SUM(myisam_bigger) as nb_myisam_bigger, SUM(mypack_bigger) as nb_mypack_bigger
    -> FROM (SELECT IF(myisam_size+myisam_size2 > initial_size, 1, 0) as myisam_bigger,
    ->              IF(mypack_size+mypack_size2 > final_size,   1, 0) as mypack_bigger
    ->       FROM compression_logs) as c;
+------------------+------------------+
| nb_myisam_bigger | nb_mypack_bigger |
+------------------+------------------+
|                0 |                0 |
+------------------+------------------+
1 row in set (0.19 sec)

I think this clearly shows that we still need MyISAM and myisampack for efficient storage of read-only tables.

A disclaimer: I am not yet using MyISAM and myisampack for the storage of this particular dataset in production.  However, I might switch from InnoDB compression to MyISAM and myisampack because, as shards are filling-up, it might be better to change the storage format than to allocate more shards/hardware for this system.