Zend Framework

Zend_Cache_Backend_Memcached bad performance on adding setting keys

Details

  • Type: Performance Improvement Performance Improvement
  • Status: Resolved Resolved
  • Priority: Major Major
  • Resolution: Fixed
  • Affects Version/s: 1.9.3
  • Fix Version/s: 1.10.0
  • Component/s: Zend_Cache
  • Labels:
    None

Description

In the Zend_Cache_Backend_Memcached->save method it tries memcache->add and if it fails does a memcache->set.
The comment just above this code says that set() seems slower.

My tests indicate that add() and set() is almost identical in speed.

Also, if done in this way we assume that the cache-keys doesn't already exist most of the time (that add() doesn't fail).
If add() fails, the saving of a cache will take double time.

I propose to just call memcache->set(), and not do add() at all. This will give consistent speed on saving the cache.

Issue Links

This issue is related to:
ZF-5702 Memcached backend performs set() calls unecessarily Major Resolved

Activity

Ascending order - Click to sort in descending order
Hide
Permalink
Marc Bennewitz (private) added a comment -

Yes, on replacing items all memcache requests are send twice.
But in inserting new one the requests are send once but memcache isn't really faster because memcache have to check if the item exits anyway.

Therefor a simple set is enough.

I'm planing to integrate some new functionality (like add, replace, multiple interface etc.) but there few toto's:
(http://framework.zend.com/wiki/display/ZFPROP/Zend_Cache+refactoring+-+Marc+Bennewitz)

Show
Marc Bennewitz (private) added a comment - Yes, on replacing items all memcache requests are send twice. But in inserting new one the requests are send once but memcache isn't really faster because memcache have to check if the item exits anyway. Therefor a simple set is enough. I'm planing to integrate some new functionality (like add, replace, multiple interface etc.) but there few toto's: (http://framework.zend.com/wiki/display/ZFPROP/Zend_Cache+refactoring+-+Marc+Bennewitz)
Hide
Permalink
Marc Bennewitz (private) added a comment -

The add method was implemented by another issue:
> save() performs a replace() when it's intending to do a set(), and replace() apparently doesn't work on older (<= 1.2.3) versions of memcached.

But I can't find any verifications for this testimony:
> set() is rather slow in memcached

Show
Marc Bennewitz (private) added a comment - The add method was implemented by another issue: > save() performs a replace() when it's intending to do a set(), and replace() apparently doesn't work on older (<= 1.2.3) versions of memcached. But I can't find any verifications for this testimony: > set() is rather slow in memcached
Hide
Permalink
Marc Bennewitz (private) added a comment -

If replace doesn't work with memcached <= 1.2.3, I don't now why replace() is used within touch() ?!??!
Can't find this replace() bug within memcached changelog. I will try to test out it within the next days.

Show
Marc Bennewitz (private) added a comment - If replace doesn't work with memcached <= 1.2.3, I don't now why replace() is used within touch() ?!??! Can't find this replace() bug within memcached changelog. I will try to test out it within the next days.
Hide
Permalink
Marc Bennewitz (private) added a comment -

I tested the replace command with memcached 1.2.3 but it works fine.

Fixed in r20590 (trunk) & r20591 (1.10 branch)

Show
Marc Bennewitz (private) added a comment - I tested the replace command with memcached 1.2.3 but it works fine. Fixed in r20590 (trunk) & r20591 (1.10 branch)

People

Vote (0)
Watch (1)

Dates

  • Created:
    Updated:
    Resolved:
Atlassian JIRA (v4.4.3#663-r165197) | Report a problem
Powered by a free Atlassian JIRA open source license for Zend Framework. Try JIRA - bug tracking software for your team.