Last Sync: 2022-12-03 12:00:04

This commit is contained in:
tactonbishop 2022-12-03 12:00:04 +00:00
parent 08b398bb57
commit 925968024a
5 changed files with 50 additions and 3 deletions

7
DevOps/AWS/AWS_Lambda.md Normal file
View file

@ -0,0 +1,7 @@
---
categories:
- DevOps
tags: [AWS]
---
# AWS Lambda

View file

@ -0,0 +1,23 @@
---
categories:
- DevOps
tags: [AWS]
---
# AWS Messaging services
## SQS
> SQS: Simple Queue Service
SQS is a service that allows you to send, store and receive messages between apps and software components built in AWS, with automatic encryption. It helps with decoupling and scaling.
As the name indicates, its operating mode is that of a [queue](/Data_Structures/Queue.md) data structure offering first-in, first-out and other queue implementations.
An example application of this would be to set up an SQS queue that receives messages and triggers a lambda whenever a new message is added.
## SNS
> SNS: Simple Notification Service
Similar to SQS but the focus is on notifications rather than messages, i.e events that fire when something specific happens, not just a message-send event. It can be used for passing notifications between applications or to persons through SMS, text, push notifications and email.

View file

@ -1,5 +1,4 @@
---
title: Image and colour encoding
categories:
- Computer Architecture
tags: [binary, binary-encoding]

View file

@ -52,6 +52,24 @@ $$
0100
$$
Which is 4.
Which is 4. This means the calculation above would be identical whether we were calculating $7 + -3$ or $7 + 13$.
The ease by which we conduct signed arithmetic with standard hardware contrasts with alternative approaches to signing numbers. An example of another approach is **signed magnitude representation**. A basic implemetation of this would be to say that for a given bit-length (6, 16, 32...) if the [most significant bit](/Electronics/Digital_Circuits/Half_adder_and_full_adder.md#binary-arithmetic)
The ease by which we conduct signed arithmetic with standard hardware contrasts with alternative approaches to signing numbers. An example of another approach is **signed magnitude representation**. A basic implemetation of this would be to say that for a given bit-length (6, 16, 32...) if the [most significant bit](/Electronics/Digital_Circuits/Half_adder_and_full_adder.md#binary-arithmetic) is a 0 then the number is positive. If it is 1 then it is negative.
This works but it requires extra complexity to in a system's design to account for the bit that has a special meaning. Adder components would need to be modified to account for it.
## Shorthand for deriving two's complement
A simple way to work out the value of a signed number as contrasted with an unsigned number is to schematize it as follows: _the most significant place has a weight equal to the negative value of that place, and all other places have weights equal to the positive values of those places_.
Thus for a 4-bit number:
// INSERT PLACE VALUE DIAGRAM HERE
Then if we add the decimal equivalents of the place value together, we get our signed number. So in the case of $-3$:
// INSERT DIAGRAM HERE
## Considerations
A limitation of signed numbers via two's complement is that it reduces the total informational capacity of a 4-bit number. Instead 16 permutations of bits giving you sixteen integers you instead have 8 integers and 8 of their negative equivalents. So if you wanted to represent integers greater than decimal 8 you would need to increase the bit length.