3 ms·
I tend to agree with you (not on the basis of facts, but on hunch). One thing about your calculations I would challenge is that 500,000 requests per day is 1,0
by hoodoof 13y ago
I tend to agree with you (not on the basis of facts, but on hunch).
One thing about your calculations I would challenge is that 500,000 requests per day is 1,000 rps. If a site gets 500,000 requests because it was on TechCrunch for example, then the traffic flow is likely to be lumpy rather than smooth, with an initial very high peak requests per second declining over time.
The point being that you can't plan for traffic by dividing requests per day into requests per second. You need to plan for peak requests in any given second in that day, which is some number up to the total, hard to work out what.
- bhauer 13y agoYes, real-world traffic will definitely vary over time. 1,000 rps is significantly beyond the average rps needed for 500,000 requests per day (which works out to an average/sustained ~6 rps). I consider the capacity to handle 1,000 rps well beyond the realm of handling the unpredictable burstiness of a 6 rps sustained load. But just to contribute some anecdotal evidence: I've never observed greater than ~50 rps from the HN #1 position. Real world traffic is not smooth, but it's also not bursty to quite the degree that would be necessary to cause the server to wince. All of this remains predicated on Amazon not throttling the CPU capacity to true zero or something approximating true zero.