I thought this guy was being sarcastic the first time I read it. If he is serious, there are so many problems with it.<p>- Getting 823 requests/second should more than achievable with one server and memcache. Nginx can serve over 2000 requests/second in a single process on a single machine in 10Mb or so of memory.<p>- his benchmark only tests one page, the first one (with 10 posts on it)<p>- Cassandra is designed for high write bandwidth, which typically wouldn't be a single author blog using an external commenting system
It is so wrong on many levels I don't even know where to start. He is scaling a blog engine (!) with a "scalability daemon" instead of generating html pages and storing them in cache. He could then serve ten thousands of pages with one server instead of 823 request of seconds and the complex setup.
Please don't use it in production, like it was already said, it's wrong on way too many levels:<p>1) This "daemon" reloads nginx (by sending "HUP" signal) every X seconds. There is nothing "dynamic" about it. Also, nginx starts new workers on each reload, but old workers are kept alive until they complete serving all requests. This means that you could end up with thousands of workers if you've got connections that take a while to complete (big files, comet servers, etc).<p>2) There is already nginx module that does exactly that (dynamically changes upstream status, starts and stops backends on demand, depending on load, etc).<p>3) nginx's cache can scale blogs much better on a single machine, there is no way to start another backend servers.