Performance in Prestashop 1.7: new improvement

I) Performance improvements in

With the release of PrestaShop, we have tried to remove a few point of contention which were affecting badly overall PrestaShop performance:

  • Specific price computation
  • Bad sale page pagination, which loaded too many products on one page
Let's check the specific price improvement!

II) Why improving specificPrice is important for performance?

According to our internal software, specificPrice computation was particularly slow, AND present across a lot of pages.

Basically our software crawls a website, profiles each page it finds, then generates a global report. SpecificPrice was listed in the top 5 most resource and time consuming operation, overall.

III) When can I expect improvements?

Each time a price is displayed, the specificPrice method is called. The logic can be greatly optimized when there is no "global" specific price rule. A global rule means it applies to all your products. Luckily, global rules don't occur that often in a real shop. In this case :

  • if no specific price at all exists for a given product, don't try to compute the specificPrice (makes sense, doesn't it?)
  • if a specific price exists for this product, the SQL query will be optimised and executed faster.

This also allows to improve the local specificPrice cache, and greatly reduces the number of SQL queries associated with it.

IV) Let's bench!

1) The configuration

To do this benchmark, we chose to use a dedicated Mini HG server from OVH, which offers high performances for a fairly decent price :
  • CPU : Intel(R) Xeon(R) CPU E5-2650 v2 @ 2.60GHz (8 cores, 16 threads)
  • Memory : 64GB
  • Disk : 2x SSD 600GB
  • RAID Controller : LSI MegaRAID SAS 9271-4i CacheVault 1G FastPath
We used a Debian Wheezy, with the standard, dotdeb & packages, and a properly optimized configuration :
  • Apache 2.4.23, with mpm_event and the following configuration :
ServerLimit             16
MaxClients              400
StartServers            3
ThreadLimit             64
ThreadsPerChild         25
MaxRequestWorkers       400
MaxConnectionsPerChild  0
  • PHP 7.0.23 & APCu + Zend OpCache v7.0.23 PHP-FPM mode with the following configuration :
pm =                    dynamic
pm.max_children       = 400
pm.start_servers      = 75
pm.min_spare_servers  = 50
pm.max_spare_servers  = 100
  •  MariaDB 10.2.8, with the following configuration :
performance_schema                    = OFF
innodb_lock_wait_timeout              = 120
default_storage_engine                = InnoDB
max_allowed_packet                    = 16M
key_buffer_size                       = 128M
table_cache                           = 4096
read_buffer_size                      = 2M
read_rnd_buffer_size                  = 1M
thread_cache_size                     = 80
join_buffer_size                      = 2M
sort_buffer_size                      = 2M
max_connections                       = 400
tmp_table_size                        = 32M
max_heap_table_size                   = 32M
table_definition_cache                = 8192
query_cache_size                      = 32M
innodb_buffer_pool_size               = 3G
innodb_log_file_size                  = 128M
innodb_log_files_in_group             = 2
innodb_log_buffer_size                = 8M
innodb_flush_log_at_trx_commit        = 2
innodb_thread_concurrency             = 64
innodb_read_io_threads                = 8
innodb_write_io_threads               = 8
innodb_flush_method                   = O_DIRECT
innodb_file_per_table                 = 1
innodb_io_capacity                    = 2000
innodb_file_format                    = Barracuda
innodb_checksum_algorithm             = crc32
innodb_log_compressed_pages           = 0
innodb_flush_neighbors                = 0
# Mutex contention
innodb_buffer_pool_instances          = 32
We bench a real shop with :
  • 1827 categories
  • 10843 cart rules
  • 1904 products
  • 2M product attributes
  • 16.5M attribute combinations
  • 57 specific prices
Memcached is used, configured locally, with 128MB of memory. The benchmarks results are the average of 5 "siege -i -b -t 20S" (20s benchmark), on representative pages which have a direct impact on users navigation and SEO :
  • category pages
  • product pages
  • home pages
  • cart pages
  • best sales pages
  • tag search pages
  • manufacturer pages
  • new product pages
  • price drop pages
  • supplier pages
  • search pages

2) The result


With this shop with few specific prices in the database, the gain is interesting. Even at low concurrency, the improved local cache reduces the number of SQL queries, and allows 1.7 to have the same level of performances than 1.6.1.

At high concurrency, the gain is even higher because a signification point of contention on the database is removed.

V) Conclusion

The performance improvement is really interesting at high concurrency, and also lower the number of SQL queries at low concurrency. However you'll see improvement only if you don't have any global specific price rules in your database, which should be the case most of the time.

To check if you have global specific price rules, just search in your database if you have any results with id_product=0 in the specific_price table!