4.1 - Scaling (TT1 CUTOFF)

Question 1

What are the 3 ways to handle more traffic?

Answer 1

1. More servers (costly, pay per second)
2. Better code (pay people to write code, takes time)
3. Do less work with caching

Question 2

How do we get the client to do less work if it’s pushed onto them?

Answer 2

Maybe some things frequently requested can be locally cached to avoid expensive DB requests

e.g: YouTube video is super compressed from DB, client uses video card to render it

Question 3

What’s a CDN?

Answer 3

Content Distribution Network:
- fancy cache

• Now we don’t host the video, some other network of servers cache it closer to the user location to reduce latency
• main goal is to store big stuff

Question 4

How does a reflection attack work?

Answer 4

1. One attacker sends a small request to some service
2. That service acks the request, but sends it to a different server (the one to attack) and it’s a huge output

Question 5

If an error in the database occurs, how does the web server handle it when the browser is sending requests?

Answer 5

It failed, but doesn’t have to stop

• host whatever it can in absence of DB?
• send good, useful error messages

Question 6

How do you recover from an error?

Answer 6

• reconnect while loop
• report errors with a log, alert system

Question 7

In the simplest form, what is cache?

Answer 7

key-value pairs

Question 8

What HTTP methods could be cached?

Answer 8

Safe methods where it won’t change, like GETs

• Definitely not POST, meant to CHANGE things on serverside

Question 9

What’s good to cache?

Answer 9

Resource intensive things that aren’t updated very frequently

suggestions from amazon/netflix, maybe updates 1x/week

trending things (precomputed cache based on user id)

Question 10

What’s memcache?

Answer 10

Cache in memory (high performance)

Question 11

With 2 web servers, if the database is down, why could it be a bad idea to get the info there?

Answer 11

We shouldn’t assume cache is valid when the database is down, could be out of data and now we need to sync them, don’t want to screw it up more

Question 12

What to do when a the database is down?

Answer 12

• send a good error message

Question 13

What’s the best request for a DOS?

Answer 13

Something that’s a small request size, but large response size (for A1 this would be GET)

Question 14

What parts of a system can be DOS’d?

Answer 14

CPU time
Network (bandwidth, connections)
Memory

Question 15

How do we address CPU to prevent DOS’s?

Answer 15

• write better code
• add more machines (horizontal scaling) or better machines
• break up the problem, add a new layer

Question 16

How do we address Connections to prevent DOS’s?

Answer 16

• fix port saturation by:
• shorter timeouts (kick people out)
• horizontal scaling (more machines,threads)

Question 17

How do we address Memory to prevent DOS’s?

Answer 17

• really depends