blob: c60d913a4c0540ad7a76c6dc8a227e72c1b6ce2f (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
|
```
2020-10-15:s5/net/mm12/noter #6be5
```
# Sidste gang
## Konklusion af opgaver fra sidste gang
Fejl i TCP afhænger meget af hvilke pakker der bliver tabt.
## TCP wireless
TCP performance is very affected by wireless errors.
Solutions:
- Can use some special ACK which uses bitfields for acking multiple packets.
Also called *SACK*, i think.
- Can Also terminate TCP in the middle and switch over to another protocol.
Then move back to TCP at the other end of wireless.
- Modify layer 2 to work better with TCP.
- Can for example check if dupack and then drop it.
FEJL PÅ SLIDES: slide 31 "down-stream" skal være "up-stream".
# Fault tolerance
Dette er noter fra mm12.
How to handle faults in a network of services?
1. Define a list of possible errors, with likelihood.
Here errors can be prioritised based on likelihood and impact.
Also called *Fault analysis*.
## Types of failures
- Crash
- Timing
- Assertive
- Arbitrary/Byzantine
## Metrics
**Availability** is the fraction of time the system is available.
For very reliable systems this should be $1-10^5$.
If there is **replication** the availability becomes $1-(1-A)^N$, where N is the
number of servers.
## Replication
> If availability is to low, just throw more servers at it. /s
Have multiple servers.
How to split traffic between servers:
- Just give a list of servers
Means clients must do failover themselves.
Sometimes this is not desirable.
- Have servers on the same ip address.
So the client is unaware.
This makes a constraint on the network, which must to load balancing.
This could be done with a load balancer or with ARP.
- Have name servers which update depending on server availability.
TODO: Cluster framework.
|
