Téléshopping, one of the most older Magento project in France. With 41Meuros, website represents a tier of the turnover. Let’s see which technical choices has been made to provide a high performance website
Hello Gilles, maybe everyone doesn’t know you yet, could you please introduce yourself?
I’m in charge of the production center at Teleshopping. I am in charge of the operation, administration and system architectures for ERP and websites and for its subsidiaries (Euroshopping, Place des tendances and partly Direct Optic).
Two of your websites, Teleshopping and Euroshopping, are Magento-developed projects. Could you please tell us more about them?
Sure, Teleshopping’s website is among the first french Magento-developed websites, deployed in April 2009, and regularly updated since. There are approximately 1M unique visitors per month.
It allows its users to order products they have seen during the TF1 TV show. Ordering products from the catalog is done through another website.
Euroshopping’s website was deployed a little later, and there are about 100 000 unique visitors per month. It allows people to order products they have seen during TV shows on different TNT channels or on the cable.
Could you describe the general architecture within these two websites?
For Teleshopping, we have 10 front-end servers available for visitors, and one more for the back-office access
Proxy cache varnish is ahead each one of the servers. This may not be the best solution, but it allows us to regulate the number of front-ends without changing the configuration or having to ask ourselves too many questions.
Concerning the application, the source code is installed on every front-end server, there’s no replication and deliveries are hand-made.
But we share the media and skin directories on an NFS resource.
We use a memcached server with 1 port to put data in cache, and another port to put sessions in cache; this way we can flush data without any impact on the user’s sessions.
Concerning the database, we use a MySQL server which is replicated on a secondary server. All accesses are done on the main server.
Last element, the CDN. Medias are served by a CDN, which allows us to relieve at the outside our front-ends, even during the business campaigns.
Source images are put on an NFS resource available on every front-end.
For Euroshopping it’s at the most the same, except there’s no varnish, the 4 servers available are more than enough to manage the load.
How often do you deploy new versions?
It’s an average, but we do about 1 delivery in the production environment per month. Sometimes we may have more deliveries than that.
Could you tell us about your last deployment?
Every developer has its own development environment on a virtualized server, which is strictly the same as the production one.
A testing environment is available to validate the package before it’s installed on a pre-production environment. This delivery is done by my team.
Once the project manager has tested and approuved this package, it is delivered in the production environment, my team still doing it.
You are sharing the front-ends server’s contents on a NAS. What content do you exactly share?
We share the media and skin directories on our NAS, a filer NetApp. This sharing system is almost as performing as the local disk access, and its vantage is that we do not have to duplicate media files.
What made you choose this setup?
This sharing process is almost as performing as the local disk access, and its vantage is that we do not have to duplicate media files.
NFS and PHP tuning allows to manage access to the shared files with different cache levels.
We can manage the stock volume without any outage.
Could you tell us about the main problems you encountered while you were setting up this model?
We didn’t have any particular problem, only some tuning was needed for the NFS:
- We mount the NFS shared folder with the following options tcp, rw, bg, hard, suid, noatime, nodiratime, nointr, timeo=600, retrans=2, rsize=32768, wsize=32768, actimeo=30
- We have increased the PHP realpath cache to have a size of 1M and a TTL of 1 day
What benefits do you get from the source sharing between the different front-ends on a NAS?
Source sharing is useful to deliver the source code only once, it is then available for all of the front-ends; we do not need a “master” server from who to replicate the files.
Presently, the websites’s sources don’t allow us to share the whole source code, partly because of the /var/reports and /var/logs directories. For the next versions if source code allows it, we intend to share the whole source code for every front-end server.