Fatal-Vortex is a very popular web based interactive game with
over 10,000 players generating several million
requests each day. As result, we require a stable
yet high performing web server suite so we can
dedicate most of our server resources to game-calculations
and not waste precious CPU cycles to serving hundreds
of HTTP requests each second.
At the end of 2003 we wanted to switch to new
hardware and test alternatives to the current
Apache based solution, which in our view is using
too much system resources.
We started to benchmark several web server platforms
such as Apache (v1, v2) and less visible servers
like Boa or khttpd. After benchmarking 10 different
web server platforms, LiteSpeed came out on top
with the best performance.
LiteSpeed was performing up to 50% better than
our Apache solution for dynamic requests, using
PHP via FastCGI interface, while at the same time
using comparatively much less CPU and RAM resources.
We had some initial reservations about using
a closed-source solution. However, after our benchmarks
we were no longer concerned and confidently deployed
LiteSpeed across our system after several days
Before our switch to LiteSpeed we had the following
server and performance specs:
- Single Dual Intel Xeon 2.4GHz with 2GB RAM
- 10GB traffic per day of dynamic content delivered by PHP
- 1,400,000 requests per day on average
- 3,000,000 requests peaks on some days
Under above specs, Apache consumed between 10
- 20% CPU time at peek times, and needed minimum
of 10% CPU even during low peak hours.
LiteSpeed, Standard Edition, on the other hand,
needed only 0.3% to 1.5% CPU at peak time with
significantly less memory usage and better response
About 6 months of deployment with LiteSpeed Standard
Edition, we were evaluating the purchase of the
Professional Edition license and contacted LiteSpeed
support staff for the first time.
We were pleasantly surprised at the fast response
time and helpfulness of the overall support. We
not only received the help we needed but also
received tips on configuration, to further enhance
the performance of LiteSpeed.
There have been several updates since we first
started using LiteSpeed and with each revision
performance has been improved, new features introduced,
and even a few features we suggested to the development
group, were realized.
The upgrade process of LiteSpeed is extremely
fluid. All we had to do is answer some yes or
no questions and initiate a graceful restart.
After 7 months, we decided to give LiteSpeed
a chance on serving static content which was still
been served by Boa and had been for more than
2 years. To this point, LiteSpeed had been deployed
only for dynamic content.
Not yet confident of LiteSpeed's ability to serve
static content, we replaced Boa with LiteSpeed
for several days using Virtual-Host configuration
under live traffic. Internally we wondered just
how long it would be before the static traffic
overwhelmed LiteSpeed. To our surprise, it never
happened. LiteSpeed was able to hold its ground.
At that time we had an average of 3,500,000 requests
per day, peaking to 4,000,000 in some days, but
LiteSpeed took it and smiled at us by serving
both the dynamic and the static content without
Nearly 8 months after switching over to LiteSpeed
we have experienced growth with additional several
thousand requests added to our daily traffic and
LiteSpeed has been able to handle our increasing
traffic with no change in performance.
- 2,000,000 dynamic content requests per day
- 4,000,000 static content requests per day
- LiteSpeed still consumes very little CPU,
only needing 0.5% to 1.8% at peak
Request response time and everything else are
still lightning fast, just like our observed benchmark
performance in our initial evaluation of LiteSpeed.
LiteSpeed's excellent support has given us the
confidence in its commitment and the raw performance
has given us the satisfaction and assurance we
desired. We welcome the future with LiteSpeed
as our foundation and watching LiteSpeed grow