Examples of Resource Estimation

Introduction#

Now that we’ve set the foundation for resource estimation, let’s make use of our background knowledge to estimate resources like servers, storage, and bandwidth. Below, we consider a scenario and a service, make assumptions, and based on those assumptions, we make estimations. Let’s jump right in!

Number of servers required#

Let’s make the following assumptions about a Twitter-like service.

Assumptions:

There are 500 million (M) daily active users (DAU).
A single user makes 20 requests per day on average.
Recall that a single server can handle 8,000 RPS.

Daily active users (DAU)	500	M
Requests on average / day	20
Total requests / day	f10	Billion
Total requests / second	f115	K
Total servers required	f15

Indeed, the number above doesn’t seem right. If we only need 15 commodity servers to serve 500M daily users, then why do big services use millions of servers in a data center? The primary reason for this is that the RPS is not enough to estimate the number of servers required to provide a service. Also, we made some underlying assumptions in the calculations above. One of the assumptions was that a request is handled by one server only. In reality, requests go through to web servers that may interact with application servers that may also request data from storage servers. Each server may take a different amount of time to handle each request. Furthermore, each request may be handled differently depending upon the state of the data center, the application, and the request itself. Remember that we have a variety of servers for providing various services within a data center.

We have established that:

Finding accurate capacity estimations is challenging due to many factors—for example, each service being designed differently, having different fan-out rules and hardware, server responsibilities changing over time, etc.
At the design level, a coarse-grained estimation is appropriate because the purpose is to come up with reasonable upper bounds on the required resources. We can also use advanced methods from the queuing theory and operations research to make better estimates. However, an interview or initial design is not an appropriate time to use such a strategy. Real systems use various methods (back-of-the-envelope calculations, simulations, prototyping, and monitoring) to improve on their initial (potentially sloppy) estimates gradually.

Therefore, we approximate the number of servers by depicting how many clients a server handles on a given day:

We assume that there is a specific second in the day when all the requests of all the users arrive at the service simultaneously. We use it to get the capacity estimation for a peak load. To do better, we will need request and response distributions, which might be available at the prototyping level. We might assume that distributions follow a particular type of distribution, for example, the Poisson distribution. By using DAU as a proxy for peak load for a specific second, we have avoided difficulties finding the arrival rates of requests and responses.

Therefore, the DAU will then become the number of requests per second. We already have RPS for a server; therefore, equation (1) becomes:

Note: Our calculations are based on an approximation that might not give us a tight bound on the number of servers, but still it’s a realistic one. Therefore, we use this approach in estimating the number of servers in our design problems. Informally, the equation given above assumes that one server can handle 8,000 users per second. We use this reference in the rest of the course as well.

Storage requirements#

In this section, we attempt to understand how storage estimation is done by using Twitter as an example. We estimate the amount of storage space required by Twitter for new tweets in a year. Let’s make the following assumptions to begin with:

We have a total of 250 M daily active users.
Each user posts three tweets in a day.
Ten percent of the tweets contain images, whereas five percent of the tweets contain a video. Any tweet containing a video will not contain an image and vice versa.
Assume that an image is 200 KB and a video is 3 MB in size on average.
The tweet text and its metadata require a total of 250 Bytes of storage in the database.

Then, the following storage space will be required:

Daily active users (DAU)	250	M
Daily tweets	3
Total requests / day	f750	M
Storage required per tweet	250	B
Storage required per image	200	KB
Storage required per video	3	MB
Storage for tweets	f187.5	GB
Storage for images	f15	TB
Storage for videos	f112.5	TB
Total storage	f128	TB

Total tweets: $250 M \times 3$ = $750 \times 10^6$ tweets per day.
Storage required for tweets in one day: $750 \times 10^6 \times 250 B= 187.5\ GB$
Storage required for 10 percent images for one day: $\frac{750 \times 10^6 \times 10}{100} \times 200 \times 10^3 B= 15\ TB$
Storage required for 5 percent video content for one day: $\frac{750 \times 10^6 \times 5}{100} \times 3 \times 10^6 B= 112.5\ TB$

Bandwidth requirements#

In order to estimate the bandwidth requirements for a service, we use the following steps:

Estimate the daily amount of incoming data to the service.
Estimate the daily amount of outgoing data from the service.
Estimate the bandwidth in Gbps (gigabits per second) by dividing the incoming and outgoing data by the number of seconds in a day.

Incoming traffic: Let’s continue from our previous example of Twitter, which requires 128 TBs of storage each day. Therefore, the incoming traffic should support the following bandwidth per second:

\frac{128 \times 10^{12}}{86400} \times 8 \approx 12 Gbps

Note: We multiply by 8 in order to convert Bytes(B) into bits(b) because bandwidth is measured in bits per second.

Outgoing traffic: Assume that a single user views 50 tweets in a day. Considering the same ratio of five percent and 10 percent for videos and images, respectively, for the 50 tweets, 2.5 tweets will contain video content whereas five tweets will contain an image. Considering that there are 250 M active daily users, we come to the following estimations:

Daily active users (DAU)	250	M
Daily tweets viewed	50	per user
Tweets viewed / second	f145	K
Bandwidth required for tweets	f0.3	Gbps
Bandwidth required for images	f23.2	Gbps
Bandwidth required for videos	f174	Gbps
Total bandwidth	f197.5	Gbps

$250 M \times 50\ tweets = 12.5$ billion tweets are viewed per day
Tweets viewed per second: $\frac{12.5}{86400} ~= 145 \times 10^3$ tweets per second
Bandwidth for tweet: $145 \times 10^3 \times 250 \times 8\ bits \approx 0.3\ Gbps$ (One tweet is equal to 250 bytes).
Bandwidth of 10 percent images in tweets in a second: $145 \times 10^3 \times \frac{10}{100} \times 200 \times 10^3 \times 8\ bits = 23.2\ Gbps$
Bandwidth for 5 percent video tweets in a second: $145 \times 10^3 \times \frac{5}{100} \times 3 \times 10^6 \times 8\ bits = 174\ Gbps$

The total outgoing traffic required will be equal to: $0.3 + 23.2 + 174 \approx 197.5\ Gbps$ .

Introduction#

Number of servers required#

Estimating the Number of Servers

Storage requirements#

Estimating Storage Requirements

Bandwidth requirements#

Estimating Bandwidth Requirements