Input, Output and Storage

1 Describe how data is stored on a magnetic hard disk drive. 3

One mark for each of the following, to a maximum of 3.

• Rigid platters coated with magnetic material rotate at high speed
• A read write head on a moving arm passes over the surface
• Regions of the surface are magnetised in one direction or another to represent 1 and 0

Model answer. A magnetic hard disk drive contains one or more rigid platters coated with a magnetic material, which rotate at high speed, typically between five and fifteen thousand revolutions per minute. A read write head mounted on a moving actuator arm is positioned nanometres above the surface and can be moved radially to reach any track. To write data, the head produces a magnetic field that magnetises a tiny region of the coating in one of two directions, representing a 1 or a 0, and to read it the head detects the direction of magnetisation as the surface passes beneath it.

2 Explain the difference between DRAM and SRAM and state where each is typically used. 4

One mark for each of the following, to a maximum of 4.

• DRAM stores each bit as charge in a capacitor which must be refreshed
• SRAM uses flip flops and needs no refresh, so it is faster
• DRAM is denser and cheaper, so it is used for main memory
• SRAM is more expensive and less dense, so it is used for cache

Model answer. DRAM stores each bit as an electrical charge in a tiny capacitor. Because that charge leaks away, every cell has to be refreshed thousands of times per second, and this refresh cycle both consumes power and makes access slower. SRAM stores each bit using a small arrangement of transistors forming a flip flop, which holds its value as long as power is supplied and therefore needs no refreshing, making it considerably faster to access. The trade off is density and cost: a DRAM cell needs roughly one transistor and one capacitor while an SRAM cell needs about six transistors, so DRAM packs far more capacity into the same area at a much lower price. DRAM is therefore used for main memory where large capacity is essential, while SRAM is used for the small quantities of cache inside and beside the processor where speed matters far more than cost.

3 Explain what wear levelling is and why it is necessary in solid state storage. 3

One mark for each of the following, to a maximum of 3.

• Flash memory cells tolerate only a finite number of program and erase cycles
• Without intervention frequently written areas would fail long before the rest
• Wear levelling distributes writes evenly across all cells to extend the drive's life

Model answer. Each cell in flash memory can only be programmed and erased a finite number of times before it becomes unreliable, typically in the order of thousands of cycles for consumer drives. Without intervention this would be a serious problem, because file systems repeatedly rewrite certain areas such as directory structures and log files, so those particular cells would wear out and fail while the vast majority of the drive remained almost unused. Wear levelling is a technique implemented by the drive's controller that tracks how many times each block has been written and deliberately redistributes data so that writes are spread evenly across every cell in the device. This means the whole drive ages at approximately the same rate, which extends its usable life very substantially.

4 A company must choose storage for a database that is accessed thousands of times per second. Justify a suitable choice. 4

One mark for each of the following, to a maximum of 4.

• Recommends solid state storage
• Access time is uniform because there is no seek or rotational delay
• Random access performance matters most for a database with many small reads
• Acknowledges the higher cost per byte as the accepted trade off

Model answer. The company should use solid state storage, ideally NVMe drives. A database serving thousands of requests per second performs a very large number of small random reads scattered across the data, and this is precisely the workload magnetic storage handles worst, because every request requires the actuator arm to seek to the correct track and then wait for the platter to rotate the sector under the head, adding several milliseconds of latency to each access. Solid state storage has no moving parts, so access time is essentially uniform regardless of where the data sits and random reads are served hundreds of times faster, which translates directly into the number of queries the system can serve. The trade off is cost per byte, which is significantly higher than magnetic storage, but for a workload where response time determines whether the service is usable, that cost is easily justified, and the capacity required for a transactional database is generally modest compared with archival data.

5 Explain what virtual memory is and why heavy use of it degrades performance. 4

One mark for each of the following, to a maximum of 4.

• Virtual memory is an area of secondary storage used as an extension of RAM
• Pages not recently used are written out to free physical memory
• They must be read back when needed, which requires further paging
• Secondary storage is orders of magnitude slower than RAM, so the processor waits

Model answer. Virtual memory is a region of secondary storage, known as the page file or swap space, which the operating system uses as an extension of physical RAM so that the total memory available to running processes can exceed the RAM actually installed. When physical memory is exhausted, the operating system selects pages that have not been accessed recently and writes them out to the page file, freeing those physical frames for other use. Performance suffers because those pages will usually be needed again, and retrieving them requires reading from secondary storage and typically evicting something else to make room. Secondary storage is orders of magnitude slower than RAM even when it is solid state, so every page fault stalls the process for a period during which the processor has nothing useful to do. When physical memory is very badly oversubscribed the system reaches a state called thrashing, in which almost all processor time is spent moving pages back and forth and almost none is spent executing the programs themselves.