Page views/uniques http://i.imgur.com/vPnEK.png
Yeah GAE might work better but they're a bit out of reach mostly due to the technologies that you are required to use. Any migration at that scale is a seriously massive matter.
I guess the funny thing is that you can't plan to grow that big. Usually it just happens suddenly (facebook/reddit style) and then you've got to make some tough architectural calls.
It's funny how folks are like "That's a problem I'd like to have..". I'm not sure they're very happy with this problem at this very moment.. sigh
In this respect, application partitioning with a deployable schema is not a bad solution for what I do.
I often hear that PHP + varnish is only limited by the size of the pipe. Can this scale on cloud platforms?
You've been brainwashed by the idiotic pop culture language wars that pretend a question like Ruby vs Python vs whatever has any meaning. These are debates over buzzwords, not reality.
All that matters is the database configuration and associated architecture. Not some buzzword drama.
4 years ago, stupid comments were brutally discouraged. People with stupid things to say would be dogpiled by negative comments.
It is pop culture that thinks asking stupid questions is a virtue.
Most people are just too tactful to tell the truth.
Also performance has a massive effect on how and what language a site is written in.
My comment is sarcastic and abstract. A literal interpretation will not suffice to understand my meaning.
If you want it literally: the language this is written in is irrelevant, because the performance differences between the various languages are negligible next to the performance differences that come about due to the precise design of the website.
A poorly engineered website in a "fast" language will still be slower than a well-engineered website in a "slow" language. The actual language is completely irrelevant next to the implementation details.
>Also performance has a massive effect on how and what language a site is written in.
Not really. Experienced engineers understand that language is basically irrelevant for a situation like this. Whether it is written in Python or something else is irrelevant information.
Additionally, newbies tend to focus on "language" and "language wars" and the comparison between languages not for any engineering or technical reason. Newbies do this because languages represent buzzwords around which various communities congregate. The Python vs. Ruby fight is as irrelevant as the Presbyterian vs. Baptist fight. It is the psychological appeal of "identity politics" that motivates people to engage in these meaningless debates about buzzwords.
Also, if you were to use Varnish as a cache, that wouldn't help you much with write access because every time there is a write (a new post..) you'd have to invalidate the cache. In a high volume scenario such as this (tons of posts coming in..), the page is as good as dynamic, even if you DO cache it for a bit.
No, PHP+Varnish can not always scale. You're forgeting the most important important factor in this equation, the database. Varnish is simply for caching and you won't find any large site that doesn't have it. PHP is a scripting language and the runtime performance differences to other common web application languages (i.e. Python, Java) is negligible since you're using caching anyway. Here's what you should worry about if you want scaling: Your database, your data structures and your code quality.
I would've thought that they'd have brought in some additional computing power for such an event, should've been easy for them considering they have a cloud deployment. Maybe this gives them greater reason to hire more engineers. I found it impressive that they served billions of impressions with just 2 engineers a short while ago..
Also, it says a lot about the "Come Cloud with us, we'll help you scale" marketing bandwagon. We've seen time and again issues with EC2s infrastructure and if EC2 doesn't have issues right now (http://status.aws.amazon.com/) then it's just sad that they can't order a gazillion instances for this event and have it scale easily.
Definitely makes me think that we still have a long way to go to compute in a truly 'elastic' way.
reddit definitely does have some crazy infrastructure in place but this would've been one of the most important moments in reddit history (so far..) and I'm sad to see that their engineers are probably going to get blamed for this..
EDIT: Okay, they're back in read only mode.. I wonder how they'll hack in some write access for the AMA while keeping everything else read only. Time for some app server redeployments! Funsies!
EDIT2: And they're gone again sigh