1. Many Cores

(1) Many-Core Challenges

In this part, we are going to talk about implementing a multi-core system. When it comes to the situation that we have more than 1 core because we need to maintain coherence between more cores, the coherence traffic increases. Therefore, we will meet the following challenges,

• Bus Bottleneck: When we have more coherence traffic on the bus, the bus only allows us to do one request at a time. Then the bus will become a bottleneck.
• Off-Chip Traffic Bottleneck: when we have more cores, we also have more off-chip traffic (i.e. memory traffic). Therefore, we…

1. Memory Inconsistency Problems

(1) Consistency Vs. Coherence

Before we discuss consistency, let’s first differ it from coherence. What coherence does is that it defines the order of accesses observable by different threads to the same address. In order to make the data on the same address the same in all the caches containing the corresponding block and in the memory, we have to maintain cache coherence. To allow coherence, we have to allow cache-to-cache transfer between caches.

However, memory consistency defines the order of accesses to different addresses. If we have out-of-order reads and writes from different addresses, the reader…

1. The Introduction of Synchronization

(1) The Definition of Synchronizations

Because we can have multicores and multi-threads that are trying to work together in the same program, we will need synchronization. For example, let’s say we have two threads and what they are doing is counting the occurrences of different letters in an essay. The first thread is called thread A, which will be counting the first half of the document, and the second one is called B, which will count the second half.

1. Introduction to Cache Coherence

(1) Cache Incoherent Problem

Before we talk about cache coherence, let’s first see why do we need to have cache coherence. When we use the shared memory for multicores, we assume that we can always retrieve the latest value from the shared memory. However, this is not the case in reality because each core has its own cache. If one core writes to its cache, and the cache has not written to the shared memory, then other cores will grab the wrong data from the memory if they need this value. …

1. Introduction to Multi-Processing

(1) Flynn’s Taxonomy of Parallel Machines

We can distinguish different types of parallel machines according to how many instruction streams they have and how many data streams these instructions operate on.

• Single Instruction, Single Data (SISD): logically, this type of machine executes one instruction stream (single program counter), and each operation operates on one data stream. This is really a normal uniprocessor.
• Single Instruction, Multiple Data (SIMD): this type of machine executes one instruction stream, but there is more than one data stream they operate on. These are commonly so-called vector processors. …

1. Introduction to Fault Tolerance

(1) The Definition of the Dependability

Dependability means the quality of a delivered service that justifies relying on the system to provide the service. So a dependable system is the one that provides the service in a way that makes us expect it to do provide the correct service.

(2) Two Kinds of Servies

The service we have talked about actually has two definitions.

• Specified service: what the behavior of the system should look like (the expected behavior)
• Delivered service: the actual behavior that the system provides

We can also conclude that dependability is about…

1. Introduction to Storage

(1) The Role of the Storage

The storage keeps all the files (i.e. programs, data, settings), virtual memory (because the physical memory is not enough to hold all the virtual memory).

1. An Introduction to Memory

(1) The Definition of Random Access Memory (RAM)

Random-access memory (RAM) simply refers to the fact that we can access any memory location by address without going through all the memory locations. The opposite definition is a sequential access tape, where we have to scan through the whole tape to get a memory location.

(2) Types of RAM

There are mainly two types of RAM,

• Statistic Random Access Memory (SRAM): the term “statistic” refers to the fact that SRAM retains its data while the power is supplied. So you are not going to lose any data…