# Idea to reduce the reset rate of transceiver

## 1. Background

### 1.1 Transceiver has 2 state

* Initialization
* Normal run time

### 1.2 Our problems present

* Initialization
* too long
* Normal run time
* Go to init. state when a error occured

Thus,

we need a **strategy** to handle when **error occured**

### 1.3 Summery

* Cost(time) is very high when error occured.
* *error occured* is a binary problem (0 or 1, Toss a coin)
* I have a strategy which inspired by *Saturating counter*

## 2. Saturating counter of branch predictor

### 2.1 Branch

One of branch :


There are many branch in our code we've written.

#### please note:

`if/else` and `conditon of while/for loop`
can be expression as **Binary branch** in low level of machine finally (asm-lang)

### 2.2 (Binary) Branch predictor

Many CS/EE student could listen branch predictor in the course *CPU Architecture* .

:::info
To improve efficiency of CPU,
we predict which branch will appear,
and **pre-execute** this branch,
finally save result into pipleline for executing.
:::

So,

how many probability about correct prediction in Branch predictor(Saturating counter)?

[This Paper](https://dl.acm.org/doi/10.1145/123465.123475) mentioned


and this algorithm applied in Intel Pentium CPU in 1993

Whattttttttttttttttttttttttttt?

Worst 80%, best 99.4%.

Considering there is a *Toss problem*,

minimum probility of correct prediction should > 50%

in the same paper,

also shows probility without branch predictor


Note
::: warning
Thinking what happens if prediction is incorrect?
:::

### 2.3 Saturating counter


白話文:

錯一次給你機會，錯到第n+1次之後才轉態

#### High probability could correctly predict

While (!stuff)
  ...
...
While (!stuff)
  ...
...
While (!stuff)
  ...
...

# stuff and !stuff almost does not change 

#### Low probability could correctly predict

While (stuff)
  ...
...
While (!stuff)
  ...
...
While (!stuff)
  ...
...
While (!stuff)
  ...
...

# stuff and !stuff change very random

:::info
Thank to we spent a lot of commands(time) to execute

While (stuff)
  ...

in almost of programs.

*Saturating counter* always has high probability to predict correctly
:::

## 3. My strategy for reduce the reset rate of transceiver

* Error occured is very low probility when run-time

* Last chapter case:

While (stuff)
  ...

is similar to

While (error)
  ...

## 4. n-bis Saturating counter

digraph hierarchy {

}

## Reference

[1] [CPU 的分支預測器是怎樣工作的？](https://www.zhihu.com/question/23973128)

###### tags: `TechReport` `DeWei`