What you will find about NGINX from its website is that it is a free, open source light weight HTTP web server and a reverse proxy. Nginx doesn't rely on threads to handle requests. Instead it uses a much more scalable event-driven (asynchronous) architecture. This architecture uses small, but more importantly, predictable amounts of memory under load. We have found that it is holding up to what it says. Here is how we ended up using NGINX for one of our client and have found it meeting up to our expectation.
We were on a performance testing assignment of a game application, where much of the game contents were static flash files (of course they are dynamic as they had embedded action scripts for execution in the front end) and image files of which quite many are of sizes above 500KB. Need not to mention that this is the compressed size and as such enabling compression did not offered any performance gains. These files were served by Apache, which also serves dynamic php requests.
Our initial performance tests did indicated that for about 1000 concurrent hits, the Apache server memory consumption shot up considerably beyond 2 GB and also quite many requests were timing out. Examination of the server configuration did reveal that the server is configured to serve 256 child processes and also that it was configured to use the ‘prefork’ mpm module, which limits each child process to use only one thread. This technically limits the number of concurrent requests the Apache Server can serve to 256.
We also figured out that Apache needs to use ‘prefork’ module in order to safely serve PHP requests. Though later versions of PHP are found to be working with the other mpm module (‘worker’, which can use multiple threads per child process) many still have concerns around thread safety. The 256 client process limit is also a hard limit and if that needs to be increased, the Apache server needs to be rebuilt.
With this diagnosis, the choices we had with us was to have the static content served out of a server other than Apache and just leave Apache to serve the PHP requests. We just then thought of trying out NGINX and without wasting much time, we went ahead and implemented it. NGINX was then configured to listen on port 80 and has been configured to serve the contents from the root folder of Apache. Apache server was configured to listen at a non standard port and NGINX has also been configured to act as a reverse proxy for all php requests by getting them served processed by Apache.
We performed the tests again and have found tremendous improvement. The average latency at 1000 concurrent hits stress level has come down to under 5 seconds from about 80 seconds. We could also observe that NGINX was consuming just around 10MB. NGINX, by itself does not has any limitation on the number concurrent threads and it is constrained by the limits of the server itself and other daemons running on it.
We were on a performance testing assignment of a game application, where much of the game contents were static flash files (of course they are dynamic as they had embedded action scripts for execution in the front end) and image files of which quite many are of sizes above 500KB. Need not to mention that this is the compressed size and as such enabling compression did not offered any performance gains. These files were served by Apache, which also serves dynamic php requests.
Our initial performance tests did indicated that for about 1000 concurrent hits, the Apache server memory consumption shot up considerably beyond 2 GB and also quite many requests were timing out. Examination of the server configuration did reveal that the server is configured to serve 256 child processes and also that it was configured to use the ‘prefork’ mpm module, which limits each child process to use only one thread. This technically limits the number of concurrent requests the Apache Server can serve to 256.
We also figured out that Apache needs to use ‘prefork’ module in order to safely serve PHP requests. Though later versions of PHP are found to be working with the other mpm module (‘worker’, which can use multiple threads per child process) many still have concerns around thread safety. The 256 client process limit is also a hard limit and if that needs to be increased, the Apache server needs to be rebuilt.
With this diagnosis, the choices we had with us was to have the static content served out of a server other than Apache and just leave Apache to serve the PHP requests. We just then thought of trying out NGINX and without wasting much time, we went ahead and implemented it. NGINX was then configured to listen on port 80 and has been configured to serve the contents from the root folder of Apache. Apache server was configured to listen at a non standard port and NGINX has also been configured to act as a reverse proxy for all php requests by getting them served processed by Apache.
We performed the tests again and have found tremendous improvement. The average latency at 1000 concurrent hits stress level has come down to under 5 seconds from about 80 seconds. We could also observe that NGINX was consuming just around 10MB. NGINX, by itself does not has any limitation on the number concurrent threads and it is constrained by the limits of the server itself and other daemons running on it.
No comments:
Post a Comment