3 ms·

>> One challenge has been the lack of engineering expertise to build infrastructure that scales. The ‘WhatsApp for feature phones’ company got more traffic than

>> One challenge has been the lack of engineering expertise to build infrastructure that scales. The ‘WhatsApp for feature phones’ company got more traffic than they could handle and could not scale up fast enough ... If I remember correctly, at the time Saya was being built, GAE (google app engine) was available so couldn't they have just used that to take care of the scalability issues whiles they gradually developed local/in-house expertise? I've always assumed the "easier path" would be to use GAE, Heroku or any other commercial PaaS until it becomes more economical to switch to an IaaS which is more hands-on and requires in-house devops or system admin skills.

mattmanser 9y ago
You can't just throw an app at GAE, Heroku or any other PaaS and it'll magically scale. Even putting aside you have to have design the program for that kind of scaling, use the right language and database, a lot of apps will have hard bottlenecks in them. Table locks, disk access, API access, whatever it is. That's part of the expertise he alludes to. For example, one client had a small customer base (10s of thousands), but was already having to use Premium tier in Azure because of performance problems (P2). Very seasonal business, massive usage around Xmas. The coming Xmas they knew they had a serious problem, triple the amount of users. The site would have gone down. Bit of refactoring and performance tweaking from me and instead they never really went above 20% CPU and could sit on Azure's S2 tier (and realistically could have used S1). I'm no scaling expert, but I at least knew where to look after a decade of working on real apps and reading articles about scaling on HN.