Capacity Planning and Estimation: How to design a service for 1 million users

[@whiteboard1335]

Great video! One point to note regarding the bandwidth reduction through video chunking-while this won't actually save bandwidth, it does make the system more resilient. Clients can perform parallel operations, enable progressive loading, etc. However, the QPS will likely increase due to multiple requests for the same video (which, as mentioned at the end, can be managed by a CDN).

[@Str1x87]

Great video. Thanks!

[@orkhan-1]

Thank you!

[@miroslavrehounek8877]

Great video! One thing I noticed at 22:19: shouldn't the video QPS increase by the extra requests we are now making to upload the videos in chunks?

[@orkhan-1]

Well spotted! Indeed it will increase number of QPS eventually, but considering that we can apply almost all of the measures in 28:35 to decrease incoming bandwidth we may approximate that the we'll manage to keep bandwidth at 440 Mbps or lower. Thank you for the comment!

[@razerx]

Great video. Saved for my future preference. :D

[@orkhan-1]

Thank you!

[@johnvaldez7155]

Hi at 14:55, the calculation of the # of servers are based on the QPS(RPS) and Server Capacity. Technically, how did you come up with the 100 for the server capacity? What is the based assumed parameters used for the server capacity? I get that there are research for a specs for a particular AWS EC2 instance based on its max QPS, but in this example, what are the basis for the assumption of 100? Thanks.

[@orkhan-1]

Thank you for your comment. I took QPS that on average most modern servers can handle. As I mentioned in 18:50 each case is different and precise number of QPS will depend on variety of factors

[@johnvaldez7155]

@@orkhan-1 I understand. However, how do you defined the average modt modern servers? Like in general did you defined it like 2vcpu, 4GB server is a 1 "server capacity"? Since this is a capacity planning video, I understand that this may be under or overplanned, but what is the "average" server capacity in this case? The "100" server capacity is very vague, so its hard to make a planning, otherwise we will have no choice but to perform precise values from testing but it will defeat the purpose of "planning" objective. Hope you can give light in this question.

[@orkhan-1]

​@@johnvaldez7155 It's a relative term. For an application of this scale (1 million DAU), an average server would start with 4 vCPUs and 16 GB of RAM, which is enough to handle approximately 100 QPS