10.2.1. Fully Associative Placement vs. Set Associative Mapping

If for a corresponding virtual page number the physical page is not there, there is not a proper translation of the virtual page to a physical page, I have a page fault. This means I need to load a virtual address from the secondary storage into a page frame in physical memory before I can access any data.

A page fault occurs when a program tries to access data that is not currently in physical memory. The system needs to fetch that data from secondary storage (like a hard drive), which is much slower than accessing data from RAM. This can result in delays and is considered a 'fault' because the system is 'faulting' on finding the requested data in memory.

Think of a library where you have a book request. If the book is in the library (physical memory), you get it quickly. But if the book is in storage off-site (secondary storage), someone has to drive to get it. This drives up the time it takes to retrieve what you want, like a page fault slows down a computer program.

The access times on secondary storage are very high compared to main memory. Accessing secondary storage can take millions of nanoseconds.

Accessing data in physical memory takes much less time (around 50 to 70 nanoseconds) compared to fetching it from secondary storage, which can take up to 5 million nanoseconds. This difference illustrates why minimizing page faults is crucial in system design.

Imagine going to a fast-food restaurant versus waiting at the DMV. The fast-food line moves quickly (main memory), while the DMV line crawls forward slowly (secondary storage). This waiting is analogous to the time lost during a page fault.

Page sizes should be large enough to amortize the high cost of accessing secondary storage. Typically today, page sizes range from 4 KB to 64 KB.

Larger page sizes help to fetch more data at once from secondary storage, which reduces the likelihood of needing more accesses and consequently minimizes page faults. The optimal page size balances memory efficiency and access speed.

It's like buying snacks in bulk versus one by one at a store. It's more convenient and cheaper to buy in bulk (larger pages), as you reduce the number of trips (page accesses) to the store (secondary storage).

In embedded systems, page sizes are typically lower (around 1 KB), due to resource constraints and predictable memory access patterns.

Embedded systems often operate within stricter limits on resources, making smaller page sizes a better fit. This minimizes wasted space, also known as internal fragmentation, which occurs when the last page of memory allocated is not completely used.

Consider fitting items into a suitcase. If the suitcase is too big, some space will be wasted. A smaller suitcase that fits perfectly means no wasted space—like how smaller pages reduce internal fragmentation in embedded systems.

Virtual memories typically use fully associative placement of pages in main memory, which allows better mapping compared to caches that may use set associative mapping.

In fully associative mapping, any virtual page can be brought to any physical memory location. This flexibility helps decrease page faults but requires checking all tags to find where a page is stored, which can be expensive. Conversely, caches use set associative mapping, which is less flexible but more hardware-efficient.

Imagine a parking lot where any car can park anywhere (fully associative), making it easy to find a spot but difficult to track. In contrast, a parking lot where cars can only park in designated sections (set associative) makes management easier but is less flexible.

Handling page faults in software allows for the use of smart replacement algorithms that help reduce cache misses.

When a page fault occurs, instead of relying on hardware to manage memory replacement, software can implement intelligent algorithms that decide which pages to discard, optimizing performance under different usage scenarios.

Think of an event planner deciding which activities to cut if time runs short—they can use experience and data to make the best decision rather than randomly eliminating subjects. Similarly, smart algorithms help the system decide which pages to replace.

A write-back mechanism is used for virtual memories, meaning that modifications are kept in physical memory and only written to secondary storage when necessary.

In write-back caching, updates to data occur in RAM, and the changes are pushed to secondary storage only when the data is no longer held in RAM. This approach greatly reduces the number of writes and the time spent accessing slower memory.

Consider cooking. If you keep tasting your dish while cooking (write-back), you adjust the flavor as you go, instead of serving and then needing to redo the dish if it’s not right (write-through). This approach saves time and ensures better results.

--