If a website encounters an influx of unexpected traffic for multiple weeks causing it to crash, how do they fix this and how long would it normally take? Also, how could it still not function during lower traffic times?
In: Technology
When you open a website (let’s say Reddit), your computer sends a message to the computer running the website (called a server) saying “hey, can you tell me what’s on the homepage of reddit.com?” and the server sends back a message saying what’s on the homepage of reddit.com.
If too many people do it at once, the server can get too busy trying to answer all the messages. Like that scene in Harry Potter where they get thousands of Hogwarts letters.
If you’re lucky then it will answer as many messages as it can and ignore the rest, but that means the site still doesn’t load for most people. If you’re unlucky it won’t answer any of them. That depends on a bunch of details like how many messages it tries to answer at the same time. Either way though, most people don’t get an answer.
They can fix it by upgrading their server or getting more servers. Maybe they can try tweaking some settings to make the server answer the messages more quickly.
Usually it doesn’t stay down after the traffic spike. Some reasons that it might:
* Maybe the “low traffic” Is still much higher than it used to be
* Maybe they took the server offline so they could do upgrades
* Maybe they rent their server from another company, and they only pay for a certain number of messages per month, and the traffic spike used them all up, so now the server company won’t allow anyone to see the site until the month ends or they pay more.
* Maybe the server program crashed, so now the server computer is just sitting there not doing anything and ignoring your messages because the program isn’t running, until the owner logs in and starts the program again. (Servers are normal computers, they don’t do things except by running programs)
When a page is being loaded up there’s a worker (thread) on your computer that makes a (tcp) connection and sends a request for the page. On the internet somewhere there’s another computer that has the page and it also assigns a worker (thread) just for you. This is a point of choking while milions can access a single site, that computer gets overloaded creating miliins of programs (one per client) to handle them. This is mitigated by having multiple computers in the same room or building and distributing the clients.
Then there’s the information overload, sometimes the page isn’t as simple and it needs to ask for some detail from some book (database). If too many people in the room fight over the single book that holds the data then it also locks up and you get weird traffic jam effects.
Then there’s what’s called a dns overload. When you first type in a name in your browser your computer has to check that name against an address to find it’s computer destination that is the country/town where the server is. It used to happen when too many people in an area went to the same “phone book” to search for an ip/adress it would lock that up and your computer couldn’t load a page it hasn’t seen before.
Hope this clears it up, there are even more limits and scenarios around traffic routing, but in all cases you can imagine it as car traffic jams in cities or highway intersections.
Websites are hosted by something called a Web server and the people connecting to it are called web clients.
Imagine a drive through restaruant, let’s call it McRolands, serves people that arrive in cars one by one. Each car places its order (requests a web page), McRoland cooks the webpage and gives it to the client. It can handle the normal amount of traffic coming to it fine.
Now imagine all of a sudden a bunch of coaches turn up with 80 people on board. It’s got to serve all them tasty web pages to the people on the coaches while the usual amount of traffic is still coming to the restaurant. There’s a big hold up, people can’t get their BigWebs and WebShakes.
McRoland has a big back order of requests that it needs to serve and people aren’t getting their stuff. Annoyed, some of the people drive off because they can’t get their stuff. Meanwhile McRoland might still be trying to process those requests so it’s still trying to serve up a Royale with Java, walking round the car park trying to find some one who’s already left.
Sometimes the server will eventually catch up as the timeouts expire on the requests and they start throwing uncollected orders in the bin. Sometimes it may just crash the whole system and requires the Area Manager to step in and turn the restaurant off and on again.
move the site to a server with more resources and/or optimize the server to handle more traffic. shouldn’t take more than a few hours.
Latest Answers