Practice · File Systems, Disks and Storage

Q1

In a Unix file system, what does a directory entry contain?

A A name and an inode number B The file's permissions and size C The file's data D The disk's bitmap

Q2

File a has a hard link b and a symbolic link c. You delete a. What happens?

A b and c both still work. B Both are broken. C c works; b is broken. D b still works (the inode's link count drops from 2 to 1); c is broken because it stores the path "a".

Q3

Which allocation method makes random access to block i of a large file slow?

A Contiguous B Linked allocation C Extents D Indexed allocation

Q4

Blocks are 4 KiB and pointers are 4 bytes. How many blocks can a single indirect block address?

A 256 B 4,096 C 1,048,576 D 1,024

Q5

An inode has 12 direct pointers, 4 KiB blocks and 4-byte pointers. The inode is cached in memory. How many disk reads are needed to read the byte at offset 49,152?

A 3 B 4 C 2 D 1

Q6

Same inode (12 direct pointers, 4 KiB blocks, 4-byte pointers, single/double/triple indirect). What is the approximate maximum file size?

A 16 GiB B 4 PiB C 4 GiB D 4 TiB

Q7

A 1 TiB disk uses 4 KiB blocks. How large is a free-space bitmap?

A 256 KiB B 256 MiB C 32 MiB D 1 GiB

Q8

A program calls write() and gets a success result; one second later the power fails. With default Linux settings, is the data on disk?

A Not necessarily: write() usually only copies data into the page cache (write-back); it is guaranteed durable only after fsync() succeeds. B No, write() never reaches the disk. C Yes, always. D Only if the file is small.

Q9

Why do databases use group commit?

A To amortize the cost of one fsync over many transactions committed together, increasing throughput B To avoid journaling C To encrypt the log D To compress data

Q10

How does a journaling file system recover after a crash?

A It reformats the disk. B It asks the user which files to keep. C It replays committed transactions from the journal and discards incomplete ones. D It scans every block of the disk.

Q11

What makes snapshots almost free in copy-on-write file systems like ZFS and Btrfs?

A Live blocks are never overwritten; a snapshot just keeps the old root pointer, and new writes go to new blocks. B They disable caching. C They compress data. D They copy the whole disk.

Q12

A 7,200 RPM disk: what is the average rotational latency?

A 0.12 ms B 4.17 ms C 8.33 ms D 7.2 ms

Q13

Average seek 9 ms, 7,200 RPM, negligible transfer time. About how many random I/Os per second can the disk do?

A 76,000 B 76 C 760 D 7.6

Q14

Head at cylinder 53; requests 98, 183, 37, 122, 14, 124, 65, 67. What is the total head movement with FCFS?

A 640 B 236 C 382 D 208

Q15

Same requests with SSTF (head at 53). In which order are they served?

A 37, 14, 65, 67, 98, 122, 124, 183 B 65, 67, 98, 122, 124, 183, 37, 14 C 98, 122, 124, 183, 65, 67, 37, 14 D 65, 67, 37, 14, 98, 122, 124, 183

Q16

What is the main weakness of SSTF?

A It moves the head more than FCFS. B Requests far from the head can starve if new nearby requests keep arriving. C It only works on SSDs. D It needs to know future requests.

Q17

Why do C-SCAN and C-LOOK exist, although they move the head more than SCAN and LOOK?

A They reduce seek time on SSDs. B They avoid rotational latency. C They are simpler to implement. D They give more uniform waiting times: requests at both ends do not wait for two full sweeps.

Q18

Which Linux I/O scheduler is typically best for a fast NVMe SSD, and why?

A SCAN, because it minimizes seek time B FCFS with a single queue C none: there is no seek to optimize, the device has many internal queues, and scheduling only adds CPU overhead D SSTF

Q19

Why can't an SSD overwrite a flash page in place?

A Flash pages must be erased before being rewritten, and erasure happens only for whole large erase blocks; the FTL therefore writes updates to fresh pages and garbage-collects. B It can; SSDs work like HDDs. C Because of RAID. D Because of the TLB.

Q20

What does the TRIM command do?

A Encrypts free space B Defragments an HDD C Tells the SSD which logical blocks no longer contain useful data, so garbage collection does not copy them D Shrinks a file

Q21

Six 4 TB disks in RAID 5. What usable capacity do you get, and how many disk failures can it survive?

A 16 TB, 2 B 12 TB, 1 per pair C 20 TB, 1 D 24 TB, 0

Q22

Why does a small random write cost 4 I/Os in RAID 5?

A Because of the journal B Because data is written to four disks C Because of TRIM D Read old data, read old parity, write new data, write new parity

Q23

Why is RAID not a backup?

A RAID loses data every day. B RAID protects against disk failures, but a deletion, corruption or ransomware encryption is immediately applied to all disks of the array. C RAID cannot store files. D RAID disks are slower.

Q24

Compare 3-way replication with Reed–Solomon RS(6, 3) erasure coding for 600 TB of data. Which statement is correct?

A Erasure coding tolerates fewer failures. B Replication stores 1,800 TB and tolerates 2 failures; RS(6,3) stores 900 TB and tolerates 3 failures, but recovering a fragment requires reading 6 fragments. C Both store 600 TB. D Replication stores 900 TB; erasure coding stores 1,800 TB.

Q25

EdgeCampus cameras produce 10 TB of video per day. Which data-placement strategy is most sensible?

A Upload all raw video to the cloud and process it there. B Delete everything immediately. C Process video at the edge, keep raw video briefly on local storage, and send only extracted results (and selected clips) to the cloud for durable storage and analytics. D Store everything only on the cameras.

Q26

What is write amplification in an SSD?

A The SSD becomes faster with more writes. B The device physically writes more data than the host requested, because garbage collection copies valid pages before erasing blocks. C The host writes each block twice for safety. D The SSD compresses data before writing it.

Q27

A process has a file open and another process deletes the file's only name with rm. What happens to the data?

A The name disappears, but the inode and data remain until the last process closes the file; then the space is freed. B rm fails. C It is erased immediately, and the reading process crashes. D The file is moved to the trash.

Q28 Short answer

Head at cylinder 53, cylinders 0–199, requests 98, 183, 37, 122, 14, 124, 65, 67. Give the service order and total head movement for SCAN and LOOK (initially moving toward 0) and for C-SCAN and C-LOOK (moving toward 199, counting the return jump). Which would you choose for a busy HDD, and why?

Q29 Short answer

A file system with 1 KiB blocks, 4-byte pointers and an inode with 10 direct pointers plus single, double and triple indirect pointers. (a) Compute the maximum file size. (b) How many disk reads are needed (inode cached) to read byte 10,240 and byte 300,000? (c) Why do modern file systems use extents?

Q30 Short answer

EdgeCampus stores sensor data on an edge server with 8 HDDs of 8 TB (150 random IOPS each). The workload is 50 % reads and 50 % small random writes. Compare RAID 5, RAID 6 and RAID 10: usable capacity, failures tolerated and effective IOPS (effective = raw IOPS / (read fraction + write fraction × write penalty)). Recommend one and justify, then say what else is needed to protect the data.