[binarycloud-dev] Zend Cache and binarycloud = more power

[Andreas A. <a.a...@th...>]

Hi,

I just benched binarycloud with/without ZendCache in comparison
with other scripts. And the results are surprising. Actually
with Zend Cache enabled I gain huge performance boosts for
sites without database access. Read on.

---------------------------------------------------------------------------
Setup
---------------------------------------------------------------------------
I ran the benchmarks on my devel box at home (crappy old box, with
lots of stuff running (sshd, samba......)):

* Intel Pentium 100
* 128MB PC100 RAM
* 15GIG IBM HDD (IDE)
* Linux Mandrake 8 (Kernel 2.4.3)
* Apache 1.3.19
* MySQL 3.23.36
* PHP4.0.4pl1 (Zend Launch Pad Edition)
* ZendOptimizer 1.1.0
* ZendCache 1.1.0

PHP compile options (launchpad - out of the box)
./configure
--prefix=/Zend
--with-apxs=/usr/local/apache/bin/apxs
--disable-debug
--with-config-file-path=/usr/local/Zend/etc
--without-db2
--disable-sysvsem
--disable-sysvshm
--enable-xml
--enable-wddx
--with-mysql
--with-gd=shared,/usr/local/gd
--enable-gd-imgstrttf
--with-jpeg-dir=/usr/local/jpeg
--with-ttf=/usr/local/freetype
--with-t1lib=/usr/local/t1lib
--with-pgsql=shared,/usr/local/postgres
--with-imap=shared,/usr/local/imap
--with-ldap=shared,/usr/local/openldap
--with-ibm-db2=shared,/usr/local/db2'
--with-oci8=shared,/usr/local/oracle

Zend Cache:
- validate timestamps is enabled (so chache checks everytime if
  script changed, will perform even better if this is disabled in
production)
- reserved cache memory: 32Megs

---------------------------------------------------------------------------
Benched scripts
---------------------------------------------------------------------------
[1] Thyrell homepage www.thyrell.de/home/index.php
A very simple script with sessions (language-detect) and smarty as template
engine (smarty cache disabled). It is just one template with basic
replacing work to do for the tpl engine (no section loops etc). No Database
access.

[2] Database script (huge)
Manage script for category administation (unlimited depth for cats).
Script is currently not r2 based [but r2 friendly and after the
entitity/database stuff is out, it will be ported:)]. MySQL Database,
about 20 entries for categories with a depth of 4.

This script is part of a more complex community application administrtion
I'm currently working on. So much things are not optimized and It's
overloaded with includes that will be removed/optimized. Escpecially
the layout for the admin interface is very grapic intensive. So currently
the loading time is very long.

Basically:
- Request class, session class of r2
- Metabase
- Smarty
- PEAR for error handling (exceptions and formwarnings)

- The "manager" responsible for db-access uses recrusion to look up
  the categories (in this example depth = 4). It also currently
  executes 2 queries per query (COUNT(*))(Few template chunks with
  descent template logic (looping over arrays).

- Advanced browser scripting (mostly IE5 behaviours).


[3] binarycloud: index.php
The current (2001-03-08, 10am GMT) cvs r2/binarycloud/build/en/index.php.
No changes.

---------------------------------------------------------------------------
The results
---------------------------------------------------------------------------
I ran the benches doubletwice and the results differ slightly.


[1] Thyrell homepage
-------------------------
With cache
Starting benchmark for http://dev.thyrell.corp/home/index.php (5 tries)...
Try 1: 6.68 req/s
Try 2: 6.90 req/s
Try 3: 6.82 req/s
Try 4: 6.79 req/s
Try 5: 6.77 req/s
6.792 requests/second from 5 tries

Without cache
Starting benchmark for http://dev.thyrell.corp/home/index.php (5 tries)...
Try 1: 0.98 req/s
Try 2: 0.60 req/s
Try 3: 0.94 req/s
Try 4: 0.86 req/s
Try 5: 0.60 req/s
0.796 requests/second from 5 tries

-> Speedup: 753.27%


[2] Database script (huge)
-------------------------------
With cache
Starting benchmark for
http://pubtour.thyrell.corp/manage/common/categories/index.php (5 tries)...
Try 1: 0.48 req/s
Try 2: 0.87 req/s
Try 3: 0.36 req/s
Try 4: 0.28 req/s
Try 5: 0.28 req/s
0.454 requests/second from 5 tries

Without cache
Starting benchmark for
http://pubtour.thyrell.corp/manage/common/categories/index.php (5 tries)...
Try 1: 0.29 req/s
Try 2: 0.28 req/s
Try 3: 0.27 req/s
Try 4: 0.27 req/s
Try 5: 0.27 req/s
0.276 requests/second from 5 tries

-> Speedup: 64.49%


[3] binarycloud
------------------
With cache
Starting benchmark for http://bc.thyrell.corp/index.php (5 tries)...
Try 1: 5.54 req/s
Try 2: 4.58 req/s
Try 3: 4.55 req/s
Try 4: 4.54 req/s
Try 5: 4.12 req/s
4.666 requests/second from 5 tries

Without cache
Starting benchmark for http://bc.thyrell.corp/index.php (5 tries)...
Try 1: 1.21 req/s
Try 2: 2.10 req/s
Try 3: 1.98 req/s
Try 4: 1.04 req/s
Try 5: 1.04 req/s
1.474 requests/second from 5 tries

-> Speedup: 216.55%



Resumee:
---------------------------------------------------------------------------
IMHO, the ZendCache is VERY groovy :-) But it's also a expensive pice of
software. So for performance critical and larger business applications it
is a must, I think. The performance boosts for non-database sites is great.
And with the binarycloud cache the pages will become real speed monsters,
I guess :-)

More detailed information is published in Zend and Techmetrix whitepapers.
You can download them from Zend.com or with some screenshots of the above
tested sites from: http://www.thyrell.de/info/downloads/cache_bench.zip.
Especially the behviour of database accesss with zend cache is interesting.
According to the paper is error free with cache (with their setup), but
there are errors without.

Andi