HPC / CFD Written Exam Preparation¶

Four practice sets (basic → very hard), short/medium/long answer templates, comparison tables, common traps, and time-management tips. Use this last week before the exam.

Time-Management Plan¶

For a typical 90-minute, 60-mark written exam:

0–5 min: read every question; mark difficulty (★ easy, ★★ medium, ★★★ hard).
5–60 min: answer in this order — easiest → hardest. Spend ~ minute per mark.
60–80 min: tackle hardest questions you skipped.
80–90 min: revise; check command syntax; write missing units.

Each answer should follow:

State (definition / claim).
Show (example / command / code / diagram).
Explain (one sentence interpretation).
Conclude (link to HPC/CFD).

If short on time, write bullet points — graders can scan them.

Answer Templates¶

Definition Question Template¶

"[Term] means [...]. It is used for [...]. In HPC/CFD, it helps because [...]. Example: [...]."

Command-Explanation Template¶

"The command cmd flags args does [...]. The flag -X means [...]. The input is [...]. The output is [...]. A common mistake is [...]."

Code-Explanation Template¶

"This code first [...]. Then [...]. Finally [...]. The output is [...]. The important concept demonstrated is [...]."

Difference-Question Template¶

"The main difference between A and B is [...]. A is used when [...]. B is used when [...]. Example: [...]."

Debugging Template¶

"The error is [...]. It happens because [...]. The corrected version is .... This works because [...]."

Long-Answer Template¶

Introduction. (1 sentence: what) Main concept. (2-3 sentences: theory) Step-by-step. (numbered list of how) Diagram. (text sketch + label) Example command/code. (block) Real HPC/CFD link. (1-2 sentences) Conclusion. (1 sentence: takeaway)

Comparison Tables (memorise)¶

Unix vs Linux¶

Aspect	Unix	Linux
Year	1970	1991
Owner	AT&T → SCO	Linus + community
Licence	proprietary	GPLv2
Cost	paid	free
Variants	AIX, Solaris, HP-UX, macOS	Ubuntu, Rocky, …

Unix vs POSIX vs GNU vs Linux¶

	Unix	POSIX	GNU	Linux
What	OS family	Standard	Userland	Kernel
When	1970	1988	1983	1991
Who	AT&T	IEEE	Stallman	Torvalds

Kernel vs User mode¶

Mode	Privileges	Crash impact	Examples
User	limited	program dies	bash, gcc
Kernel	full	system panic	drivers, schedulers

Absolute vs Relative path¶

	Absolute	Relative
Starts with `/`	yes	no
Depends on cwd	no	yes
Example	`/home/me/x`	`../data/x`

`>` `>>` `<` `2>` `2>&1`¶

Op	Action
`>`	overwrite stdout
`>>`	append stdout
`<`	feed stdin from file
`2>`	redirect stderr
`2>&1`	merge stderr into stdout

Redirection vs Pipe¶

	redirect `>`	pipe `\\|`
target	file on disk	next command
persistent	yes	no

Regex vs Globbing¶

	Regex	Glob
Used by	grep/sed/awk/Vim	shell
Operates on	text	filenames
`*` means	0+ of previous	any string
`?` means	optional previous	one char
Anchors	`^ $`	none

BRE vs ERE¶

Char	BRE	ERE
`()` `{}` `+` `?` `\\|`	literal	meta
escape gives	meta	literal
Tools	grep, sed	grep -E, awk

`grep` vs `awk` vs `sed`¶

	grep	awk	sed
Job	select lines	field/record processing	stream edit
Strength	fast filter	arithmetic	substitution
Example	`grep ERROR f`	`awk -F, '{print $1}'`	`sed 's/foo/bar/g'`

`scp` vs `rsync` vs `sftp` vs `sshfs`¶

	scp	rsync	sftp	sshfs
Mode	one-shot	one-shot incremental	interactive	mount
Resumable	no	yes	partial	n/a
Best for	quick copy	repeated big	manual exploration	edit live

`while` vs `until`¶

	while	until
Loops while	cond is true	cond is false
Exits when	cond becomes false	cond becomes true

`for` (list) vs `for` (C-style) vs `while`¶

	`for x in list`	`for ((;;))`	`while [ cond ]`
Use	known list	counter	condition
Example	`for f in *.cpp`	`for ((i=0;i<5;++i))`	`while read l`

Script vs compiled program¶

	Script	Compiled
Execution	interpreter	CPU-native
Speed	slow for math	fast
Edit-run cycle	fast	recompile

Call by value / pointer / reference¶

	value	pointer	reference
Syntax	`f(int)`	`f(int*)`	`f(int&)`
Caller passes	copy	address	original alias
Modifies caller	no	yes	yes
Nullable	n/a	yes	no

Pointer vs Reference¶

	pointer	reference
Syntax	`int p; p`	`int& r; r`
Reseat	yes	no
Null	yes	no

Git commit vs push¶

	commit	push
Where	local repo	remote repo
Visible to team	no	yes

Branch vs Merge¶

	branch	merge
Action	new line of dev	combine two
Command	`git switch -c x`	`git merge x`

OpenMP vs MPI¶

	OpenMP	MPI
Memory	shared	distributed
Within	one node	many nodes
API	pragmas	function calls
Compile	`-fopenmp`	`mpic++`

Make vs CMake¶

	Make	CMake
Type	builder	generator
Cross-platform	limited	yes
Files	Makefile	CMakeLists.txt

gdb step vs next¶

	step	next
Function call	enters	runs as one unit
Use	unknown function	known function

valgrind vs ASan¶

	valgrind	ASan
Recompile	no	yes
Slowdown	~30×	~2×

PRACTICE SET 1 — BASIC¶

Theory¶

Q1.1 Define HPC.
A. HPC aggregates many cores so calculations finish in hours not years; needed for CFD where $10^{9}$ cells × $10^{5}$ steps require $\geq 10^{15}$ ops.

Q1.2 What is the kernel?
A. The privileged core of the OS that schedules processes, manages memory, talks to drivers, and serves system calls.

Q1.3 Difference between login shell and non-login shell.
A. A login shell reads /etc/profile + ~/.bash_profile; a non-login interactive shell reads ~/.bashrc. Convention: source .bashrc from .bash_profile.

Commands¶

Q1.4 Write chmod to give owner read+write+execute, others read.
A. chmod 744 file (or symbolic chmod u=rwx,go=r file).

Q1.5 Show contents of file ~/data.txt.
A. cat ~/data.txt.

Q1.6 Find every .log file in ~.
A. find ~ -type f -name "*.log".

Output Prediction¶

Q1.7 echo "x" > a; echo "y" > a; cat a →
A. y (overwrite).

Q1.8 echo "x" >> a; echo "y" >> a; cat a (after : > a) →
A. x newline y.

Marking-Scheme Answers¶

For each: 1 mark per concept (max 3). E.g. Q1.1: 1 HPC name expansion + 1 reason for CFD + 1 example.

Common mistakes¶

Confusing > with >>.
Forgetting chmod is chmod (not chmode).

PRACTICE SET 2 — INTERMEDIATE¶

Output Prediction¶

Q2.1 Predict:

$ ls -l | grep "^d" | wc -l

A. Number of subdirectories in current directory.

Q2.2 Given data.csv:

id,name,dept,salary
1,Alice,Eng,80000
2,Bob,Sales,40000
3,Carol,Eng,95000

Run: awk -F, 'NR>1 && $3=="Eng"{s+=$4; n++} END{print s/n}' data.csv
A. 87500.

Q2.3 Given t.txt:

foo bar foo
baz foo

Run: sed -i 's/foo/FOO/' t.txt; cat t.txt →
A.

FOO bar foo
baz FOO

(only first per line because no g).

Small Bash scripts¶

Q2.4 Write a Bash script that prints "even" or "odd" for a positional argument.
A.

#!/bin/bash
[ -z "$1" ] && { echo "need arg"; exit 1; }
(( $1 % 2 == 0 )) && echo even || echo odd

Regex / grep / sed / awk¶

Q2.5 Regex matching German postal codes (5 digits).
A. ^[0-9]{5}$.

Q2.6 sed to comment lines containing TODO with //.
A. sed -E 's|^(.*TODO.*)$|// \1|' file.

Q2.7 awk: column 2 sum but only when col 3 != "skip".
A. awk '$3 != "skip" {s+=$2} END{print s}' file.

Git workflow¶

Q2.8 Steps to push a new branch feature/x and open a Merge Request.
A. git switch -c feature/x; ...edit...; git add .; git commit -m "..."; git push -u origin feature/x then open MR in GitLab UI.

Marking & Common Mistakes¶

Q2.1: count includes "total" line if not filtered. (Use ^d to be safe.)
Q2.3: forgot g flag → only one substitution per line. The example is intentional.

PRACTICE SET 3 — HARD¶

Debugging¶

Q3.1 Find the bug:

int *p; *p = 5;

A. p is uninitialised; *p writes to random memory → segfault. Fix: initialise (int a; int *p = &a; *p = 5;).

Q3.2 Find the bug:

if [ $a == "yes" ]; then echo OK; fi

A. Single brackets need spaces around = (not ==); also unquoted $a may break if empty. Fix: if [ "$a" = "yes" ]; then ... fi or [[ "$a" == "yes" ]].

Q3.3 Find the bug:

while [ $c -lt 5 ]; do
    arr[$c]=23
done

A. Infinite loop — c never increments. Fix: ((c++)) inside.

Multi-step workflows¶

Q3.4 Edit a file remotely, commit, push, and submit the result on cluster.

A.

ssh hpc
vim ~/code/solver.cpp
cd ~/code
git add solver.cpp && git commit -m "fix" && git push
sbatch run.sbatch

C++ pointer/reference¶

Q3.5 Predict:

int a = 10;
int& r = a;
int* p = &a;
r = 20;
*p = 30;
std::cout << a << " " << r << " " << *p;

A. 30 30 30 — all three refer to the same memory.

Gnuplot¶

Q3.6 Write a script that fits f(x)=a*log(x)+b to data.txt and overlays it on a scatter.

A.

f(x) = a*log(x) + b
fit f(x) 'data.txt' u 1:2 via a,b
plot 'data.txt' u 1:2 t 'data' w p, f(x) t sprintf('fit %.2f log+%.2f',a,b) w l

Shell scripting¶

Q3.7 Modify backup-data.sh to add a -d flag that deletes files in DEST not present in SRC (use --delete with rsync).

A. Add d to OPTSTRING, set DELETE=1, and pass --delete to rsync when 1.

PRACTICE SET 4 — VERY HARD¶

Mixed integration¶

Q4.1 Describe in 250 words a CFD researcher's full daily workflow, naming the relevant chapter for each step.

A. Edit (Ch. 5) ⇒ Git commit (Ch. 10) ⇒ ssh+rsync (Ch. 4) ⇒ module load + cmake build (Ch. 12 + 15) ⇒ sbatch SLURM job (Ch. 15) ⇒ MPI/OpenMP execution (Ch. 14) ⇒ awk/grep/sed result extraction (Ch. 7) ⇒ gnuplot postprocessing (Ch. 9) ⇒ rsync results to laptop (Ch. 4) ⇒ Git tag (Ch. 10) for paper reproducibility.

Real HPC/CFD¶

Q4.2 Given the following SBATCH header:

#SBATCH --nodes=8 --ntasks-per-node=4 --cpus-per-task=8 --time=06:00:00

How many cores total does the job use? What is the maximum walltime in seconds?
A. 8×4×8 = 256 cores; 6 × 3600 = 21 600 s walltime.

Q4.3 A CFD job runs at 4 PFLOPS on 1024 nodes; serial fraction estimated 1%. Predict speed-up vs perfect scaling at 16 384 nodes.
A. Amdahl with f=0.01: $S=1/(0.01+(0.99)/16384)\approx 100$; perfect would give 16 384× of single node — Amdahl caps at 100×.

Long Answer¶

Q4.4 Walk through Chapter 12's three demos (compiling, Makefile, CMake). Explain what each demonstrates and why both Make and CMake exist.

A. - demo01_compiling: shows the four-stage pipeline .cpp→.i→.s→.o→exec plus header-guard prevention of double-inclusion. - demo02_makefile: hand-written Makefile builds three shared libraries and the main executable; shows -fPIC -shared, -L/-l, and the rule syntax. - demo03_cmake: same project via CMakeLists.txt; CMake generates the Makefile, handles output directories and installation. - Why both? Make is direct but platform-specific; CMake adds portability, dependency management, and integrations (find_package, ctest, install). Modern CFD codes (SU2, OpenFOAM-extend, AMReX) use CMake or its successors.

Complex debugging¶

Q4.5 A simulation hangs after 10 hours on rank 7 only. List 5 possible causes and how you'd diagnose each.

A. (1) MPI deadlock — gstack or mpirun gdb to attach; (2) IO bottleneck on a Lustre OST — lfs check; (3) memory exhaustion — sacct MaxRSS; (4) unbalanced work — profile with score-p / valgrind; (5) network packet loss — dmesg. For each: collect logs, attach gdb, reproduce in interactive salloc.

Marking & Common mistakes¶

Q4.2: forgetting nodes × ntasks-per-node × cpus-per-task = total.
Q4.3: forgetting Amdahl plateau at $1/f$.
Q4.4: confusing Make's target/dep/recipe with CMake's imperative-vs-declarative model.

Common Written-Exam Traps¶

Trap	Right answer
`chmode`	`chmod`
`[$a==$b]`	`[ "$a" = "$b" ]`
`>` overwrites instead of appends	use `>>`
Forgot `git add` before commit	always stage first
Forgot `-E` in grep/sed for ERE	add `-E`
Forgot `-O0 -g` for gdb	optimisations remove vars
Forgot `MPI_Finalize`	always call
Forgot `module load` inside sbatch script	inherits clean env
Forgot quotes around `"$VAR"`	safety with spaces
TAB vs spaces in Makefile	TAB only
Login-node compute jobs	always submit via `sbatch`
Permission 644 for SSH private key	needs 600

Bangla Quick Notes (most-asked terms)¶

HPC = অনেক কম্পিউটার একসাথে কাজ করিয়ে বড় সমস্যা দ্রুত সমাধান করা।
Kernel = অপারেটিং সিস্টেমের মূল অংশ যা হার্ডওয়্যার নিয়ন্ত্রণ করে।
POSIX = ইউনিক্স-জাতীয় সিস্টেমের জন্য মান যা পোর্টেবিলিটি নিশ্চিত করে।
chmod 755 = মালিকের সব, অন্যদের পড়া + চালানোর অনুমতি।
ssh-keygen = পাবলিক/প্রাইভেট কী জোড়া তৈরি করার কমান্ড।
rsync = শুধু পরিবর্তিত অংশ কপি করে যা scp এর চেয়ে দ্রুত।
Vim macro = কীস্ট্রোক রেকর্ড করে পরে রিপ্লে করার ক্ষমতা।
regex = টেক্সট প্যাটার্ন বর্ণনা করার ভাষা।
awk = কলাম-ভিত্তিক টেক্সট প্রসেসিং ভাষা।
sed = লাইন-বাই-লাইন স্ট্রীম এডিটর।
gdb = ডিবাগার, ব্রেকপয়েন্ট সেট করা যায়।
valgrind = মেমরি লিক ও ভুল সনাক্তকারী।
OpenMP = থ্রেড-ভিত্তিক শেয়ারড মেমরি সমান্তরাল।
MPI = বার্তা-ভিত্তিক ডিস্ট্রিবিউটেড মেমরি সমান্তরাল।
SLURM = জব শিডিউলার যা ক্লাস্টারে কাজ বরাদ্দ করে।

German Quick Notes (most-asked terms)¶

HPC = Hochleistungsrechnen
Kernel = Kern
POSIX = POSIX-Standard
Berechtigung = Permission
Anmeldeknoten = Login node
Rechenknoten = Compute node
Auftragsskript = Job script
Verschlüsselung = Encryption
Symmetrisch / Asymmetrisch = Symmetric / Asymmetric
Pufferüberlauf = Buffer overflow
Speicherleck = Memory leak
Wettlaufbedingung = Race condition
Verklemmung = Deadlock
Übersetzer = Compiler
Linker = Linker
Statische / Dynamische Bibliothek = Static / Dynamic library
Header-Schutz = Header guard
Gebietszerlegung = Domain decomposition
Geisterzelle = Ghost / halo cell
Wandzeit = Walltime
Warteschlange = Queue
Verteilter Speicher = Distributed memory
Gemeinsamer Speicher = Shared memory

End of written-exam preparation file.

HPC / CFD Written Exam Preparation¶

Time-Management Plan¶

Answer Templates¶

Definition Question Template¶

Command-Explanation Template¶

Code-Explanation Template¶

Difference-Question Template¶

Debugging Template¶

Long-Answer Template¶

Comparison Tables (memorise)¶

Unix vs Linux¶

Unix vs POSIX vs GNU vs Linux¶

Kernel vs User mode¶

Absolute vs Relative path¶

> >> < 2> 2>&1¶

Redirection vs Pipe¶

Regex vs Globbing¶

BRE vs ERE¶

grep vs awk vs sed¶

scp vs rsync vs sftp vs sshfs¶

while vs until¶

for (list) vs for (C-style) vs while¶

Script vs compiled program¶

Call by value / pointer / reference¶

Pointer vs Reference¶

Git commit vs push¶

Branch vs Merge¶

OpenMP vs MPI¶

Make vs CMake¶

gdb step vs next¶

valgrind vs ASan¶

PRACTICE SET 1 — BASIC¶

Theory¶

Commands¶

Output Prediction¶

Marking-Scheme Answers¶

Common mistakes¶

PRACTICE SET 2 — INTERMEDIATE¶

Output Prediction¶

Small Bash scripts¶

Regex / grep / sed / awk¶

Git workflow¶

Marking & Common Mistakes¶

PRACTICE SET 3 — HARD¶

Debugging¶

Multi-step workflows¶

C++ pointer/reference¶

Gnuplot¶

Shell scripting¶

PRACTICE SET 4 — VERY HARD¶

Mixed integration¶

Real HPC/CFD¶

Long Answer¶

Complex debugging¶

Marking & Common mistakes¶

Common Written-Exam Traps¶

Bangla Quick Notes (most-asked terms)¶

German Quick Notes (most-asked terms)¶

`>` `>>` `<` `2>` `2>&1`¶

`grep` vs `awk` vs `sed`¶

`scp` vs `rsync` vs `sftp` vs `sshfs`¶

`while` vs `until`¶

`for` (list) vs `for` (C-style) vs `while`¶