A Level Computer Science Paper 1 B: Computer Systems

The program counter points to the next instruction, the MAR holds an address, the MDR holds what travels on the data bus, and the CIR holds the instruction being worked on.

Pipelining overlaps the stages of the fetch-decode-execute cycle: while one instruction executes, the next is being decoded and the one after that fetched. No part of the processor sits idle, so more instructions complete per second. A conditional branch is not resolved until it executes, by which time the instructions after it have already been fetched and decoded. If the branch is taken those instructions are the wrong ones, so the pipeline must be flushed and refilled, wasting the cycles spent on them.

CISC does more per instruction with more complex hardware; RISC keeps instructions simple and uniform, which suits pipelining.

A GPU contains thousands of simple cores designed to carry out the same instruction on many items of data at once. Tasks made of large numbers of independent, identical calculations, such as working out the colour of every pixel in an image or the matrix arithmetic in machine learning, can be split across all of those cores, so they complete far faster than on a CPU's few powerful cores.

A runs 0-2, B 2-4, C 4-6, A 6-8, B 8-9 and finishes, C 9-11 and finishes, then A 11-12 and finishes.

Shortest job first always picks the shortest process, so if short ones keep arriving a long one may never run.

Paging divides memory into equal, fixed-size pages. The divisions are physical: a page boundary can fall in the middle of a subroutine. Segmentation divides memory into segments of different sizes, and the divisions are logical, so each segment holds a complete part of the program, such as a subroutine or a data structure.

At the end of every fetch-decode-execute cycle the processor checks the interrupt register. If an interrupt is waiting and its priority is higher than the current task, the processor saves the contents of its registers, including the program counter, onto a stack. The address of the matching interrupt service routine is loaded into the program counter, so the ISR runs. When it finishes, the saved values are popped back off the stack and the interrupted task carries on exactly where it stopped. An interrupt of lower priority waits until the current task allows it.

Guaranteed response times are real-time, work spread over networked computers is distributed, many users of one machine is multi-user, and an OS built into a device is embedded.

Lexical analysis tokenises, syntax analysis checks the grammar and builds the tree, code generation produces object code and optimisation improves it.

With static linking the linker copies the library code the program uses into the executable. The file is larger, but it is self-contained and always has the version of the library it was built with. With dynamic linking the executable only holds a reference, and the library, a separate file such as a DLL, is loaded when the program runs and can be shared by many programs. Executables are smaller and pick up library updates automatically, but the program will not run if the library is missing or an incompatible version is installed.

Sequential sign-off is waterfall, repeated prototypes are RAD, risk-driven cycles are the spiral model, and pair programming with continual testing is extreme programming.

The waterfall model completes each stage, from analysis to maintenance, before the next begins, with documentation signed off at each step. That makes the project easy to plan and manage, and the thorough documentation helps a bank show regulators how the app was designed and tested. Its weakness here is that the requirements are fixed at the start: the client expects them to change as customers try the app, and in waterfall going back to change the design is slow and expensive. The client also sees a working product only near the end, so a misunderstanding may be discovered when it is most costly to fix. An agile methodology such as extreme programming builds the app in short iterations, each producing a working version that the client and real customers can try. Their feedback shapes the next iteration, so changing requirements are expected rather than a problem. Extreme programming also suits the need for security and reliability: programmers work in pairs, so every line is reviewed as it is written, and tests are written first and run continually, so faults are caught early and the code is refactored to stay clean. The drawbacks are that agile projects produce less documentation, which a bank may need for compliance, the client must stay closely involved throughout, and the final cost and delivery date are harder to predict. Overall an agile methodology is more suitable, because the requirements will certainly change, but the team should keep enough documentation of its security design and testing to satisfy the regulators.

73 is 0100 1001. In two's complement, invert to 1011 0110 and add 1: 1011 0111. In sign and magnitude, set the sign bit: 1100 1001.

The exponent 0011 is 3, so the binary point moves 3 places right: 0101.1000, which is 4 + 1 + 0.5 = 5.5.

The mantissa must start 0.1, so shift it 2 places left to 0.1101000 and reduce the exponent by 2, from 2 to 0: 0000. Both represent 0.8125.

The exponent 0010 is 2, so the binary point moves 2 places right: 101.10000. In two's complement that is -4 + 1 + 0.5 = -2.5.

AND with 0000 1111 keeps only the low four bits: 0000 0110. OR with 1000 0000 sets the top bit: 1011 0101. XOR with 1111 1111 flips every bit: 0101 0101.

A normalised mantissa has no wasted leading bits, so every bit of the mantissa adds precision. It also means each number can be written in only one way, which makes numbers easy to compare.

1000 0000 is -128 and 0111 1111 is 127.

After push(5), push(9) and pop() the stack holds 5. push(3) and push(7) give 5, 3, 7, and pop() removes 7. push(1) gives 5, 3, 1 from bottom to top.

The items are in positions 4, 5, 0 and 1, wrapping round: 4 items.

50 is the root, with 30 and 70 as its children, 20 and 40 under 30, 60 and 80 under 70, and 35 as the left child of 40. Pre-order visits the root first, in-order visits it between its subtrees, giving the values in order, and post-order visits it last.

50, then 30, then 40, then 35: four comparisons.

A hashing function is applied to the record's key to calculate its position in the table, and the record is stored there. To find it again, the same function is applied to the key, which leads straight to the right position without searching. If two keys hash to the same position, a collision, the second record can be put in the next free position, which is then searched in order, or each position can hold a linked list of every record that hashes to it.

Cherry belongs between Banana and Fig. Cherry takes Banana's old pointer, 2, so it points to Fig, and Banana's pointer changes to 5, Cherry's position.

The most recent call must return first, which is last in, first out: a stack.

A ∧ (B ∨ ¬B) = A ∧ 1 = A.

De Morgan's law: negate each term and change OR to AND.

Every cell where C is 1 is 1. A ∧ ¬B adds the cell where AB is 10 and C is 0.

The row AB = 11 is all 1s, giving A ∧ B, and the column CD = 01 is all 1s, giving ¬C ∧ D.

Sum is 1 when exactly one input is 1. Carry is 1 only when both are 1.

Sum = A XOR B and Carry = A AND B.

A D-type flip-flop stores one bit, so it is a building block of registers and memory. On each rising edge of the clock, the output Q takes the value of the data input D. Between rising edges Q holds that value, however D changes.

Each table's own ID is its primary key. OwnerID in Pet and PetID in Visit link to other tables, so they are foreign keys.

ORDER BY sorts in ascending order unless DESC is given.

SELECT Name, Species
FROM Pet
WHERE OwnerID = 12
ORDER BY Name

The join condition pairs each pet with its own owner. An INNER JOIN ... ON Owner.OwnerID = Pet.OwnerID with WHERE Species = 'Rabbit' is equally correct.

SELECT Surname, Name
FROM Owner, Pet
WHERE Owner.OwnerID = Pet.OwnerID
AND Species = 'Rabbit'

Visit 108 cost exactly £20.00, so it must stay: the condition is < 20, not <= 20.

DELETE FROM Visit
WHERE Cost < 20

When the first user opens the record to edit it, the record is locked. A second user cannot change it until the first user's transaction has been completed and the lock released. The second user then works on the updated record, so their change is added to the first rather than overwriting it.

Application formats, transport packetises and numbers, internet addresses and routes, link transmits.

The browser first checks its local cache in case the address was looked up recently. If not, it asks a DNS server. If that server does not hold the record, the query is passed up the hierarchy: a root server points to the server for the top level domain, such as .uk, which points to the domain's authoritative server, which holds the IP address. The address is returned to the browser, cached, and used to connect to the web server.

Circuit switching sets up a dedicated path between the two ends for the whole communication, so data arrives in order with a constant delay, which suits real-time voice, but the path is reserved even when nothing is being sent. Packet switching splits data into numbered packets that each take the best route available, so the links are shared efficiently and packets can be rerouted around a failure. The packets may arrive out of order, and are reassembled using their sequence numbers.

2 / 4 + 1 / 1 = 1.5. 0.85 x 1.5 = 1.275. 0.15 + 1.275 = 1.425.

Dictionary coding builds a dictionary of the words and patterns that appear in the file. Each occurrence is replaced by its short index in the dictionary, which takes far fewer bits than the word itself. The dictionary is stored with the file, so the original can be rebuilt exactly: the compression is lossless.

The shop publishes its public key and keeps the matching private key secret. The customer's browser encrypts the card details with the shop's public key. Data encrypted with that key can only be decrypted with the private key, which only the shop holds, so anyone intercepting the data in transit cannot read it.

Hashing is one-way, so even if attackers steal the database they cannot turn the hashes back into passwords. The site never needs the password itself: it hashes whatever is typed at login and checks that it matches the stored hash.

Supporters argue that live facial recognition helps the police find wanted criminals and missing people quickly, and that its presence may deter crime, making the city centre safer. Legally, a face is biometric data, which the Data Protection Act 2018 and UK GDPR treat as special category personal data. The police need a clear lawful basis, must show the use is necessary and proportionate, must delete the images of people who do not match straight away and must keep the data secure. Covert surveillance by public bodies is also governed by the Regulation of Investigatory Powers Act 2000, and anyone who hacked the system or leaked the watch list would commit an offence under the Computer Misuse Act 1990. Morally, everyone who walks past is scanned without their consent, though nearly all of them are innocent, which is a serious intrusion on privacy. No system is perfectly accurate: a false match can lead to an innocent person being stopped and searched in public. Systems trained on unrepresentative data have been shown to be less accurate for some ethnic groups and for women, so the harm of false matches falls unequally, which is discriminatory. Culturally, people who know they are being watched may change their behaviour, for example avoiding a lawful protest, which has a chilling effect on freedom of expression and assembly and can damage trust between the public and the police. There is also the risk of function creep, in which a system introduced to find wanted criminals is gradually extended to track the movements of ordinary people. On balance the technology can be justified only with strong safeguards: a narrow watch list, a trained officer checking every match before anyone is stopped, clear signs telling the public it is in use, independent oversight and published accuracy figures. Without those, the harm to privacy and fairness outweighs the benefit.