2010/1/11 Sad Clouds <firstname.lastname@example.org>
> On Monday 11 January 2010 20:37:20 Mark Robson wrote:
> > All of the focus of performance and reliability work needs to be at the
> > middle / back end. This is where all the world of pain begins.
> OK if you don't mind can you give some details where the bottlenecks are on
> your site. If you started getting a lot more web traffic what component do
> think would affect the performance?
I'm pretty sure it would be the database.
All of the things you mentioned are to do with front-end efficiency. We
don't worry about that; our front end servers also do SSL (which is itself
rather inefficient). We have only done a small amount of optimisation of the
The problems are all around a single central database, which we have
(although it has high availability components) because it's simply much
easier to manage a consistent state.
Also the numerous other databases which store large data in various forms
which we are often refactoring etc.
The amount of hardware required for web front ends is quite minimal; we have
only two boxes active (in RR DNS) for our main portal, and one could easily
handle the load (There are a number of spares which could be brought online
at short notice and are used occasionally for operational reasons). The
hardware is commodity rack-mount servers, with standard internal (RAID,
SCSI) discs and only moderate (6-16G) amounts of memory and a few cores (4-8
This message was posted to the following mailing lists: