[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Open source load balancers fail on opening day digital museum

Hello, Gijs
I think the article title is somewhat misleading.
The site failed because of traffic underestimation / unexpected, and was
using virtual machine servers with sw load balancers, all services
hosted at same machines.
Apache, Tomcat, PostgreSQL and Solr at same machine...
Even with 2 more hw servers it was under severe pressure.
It was not using massive traffic coping site planning. It was not
designed for scalability, only for functionality.
So, the failure was not of open source load balancers, but due to a
design not suitable for such traffic.
Also, the site did not crash even at such load, but slowed almost to a
halt, using minutes for each navigation event.
So, the sw load balancers seemed to be reliable (by the texts [0] [1]) ,
but at such virtualized environment, the performance was not enough.
It was tested for 5 million users/hour and received 10 millions [1], 4k
users simultaneously.
Apache slows at swapping.
The article title could be changed into something that shows the
unexpected site success.
I hope the architecture redesign could survive the massive traffic.
The "principle" article section states the advantages of using FLOSS for
the project and could be leveraged.

Please, keep us informed about the new upcoming site architecture and
the site relaunch.
It may become a good Debian success story.
Andre Felipe


Reply to: