DSy - Distributed Systems FS2025
UPDATED ACTUAL Exam Questions and
CORRECT Answers
Why do we need distributed systems? - CORRECT ANSWER - Scaling (if one machine is
not enough), Location (to move closer to the user), Fault-Tolerance (HW will eventually fail)
Vertical Scaling - CORRECT ANSWER - more memory, faster CPU (scale up)
Horizontal Scaling - CORRECT ANSWER - more machines (scale out)
Scaling for Machine Learning - CORRECT ANSWER - Current Trend: Scale horizontally,
Building models, fine Tuning → faster with multiple cards (1 User = 1 Card), storing data →
scalable storage, ML with vertical scaling is not possible
Pros / Cons Horizontal Scaling - CORRECT ANSWER - + Lower cost with massive scale
+ Easier to add fault-tolerance
+ Higher availability
− Adaption of software required
− More complex system, more components involved
Pros / Cons Vertical Scaling - CORRECT ANSWER - + Lower cost with small scale
+ No adaption of software required
+ Less complexity
− HW limits for scaling
− Risk of HW failure causing outage
− More difficult to add fault-tolerance
, Moore's Law - CORRECT ANSWER - Nr. of transistors on microchips doubles every two
years (other predictions, doubling chip performance every 18 month)
Nielsen's Law - CORRECT ANSWER - a high-end user's connection speed grows by 50%
per year, bandwidth grows slower than computer power → telecom companies are conservative,
user are reluctant to spend much money on bandwidth, the user base is getting broader
Kryder's Law - CORRECT ANSWER - disk density doubling every 13 months
Latency - CORRECT ANSWER - time for a signal to travel from source to destination and
back (RTT - round-trip time)
Why should one place a service closer to a user? - CORRECT ANSWER - Reduced
latency, can increase bandwidth and throughput, can improve reliability and availability,
drawback: coordination of data replication and caching
CDN - Content Delivery Network - CORRECT ANSWER - also: Content distribution
network, distributed databases, edge computing, placing images, sites, scripts, close to the
respective user (physically)
Why do random bit flips in memory occur? - CORRECT ANSWER - Bad Pin
connections, incorrect RAM timings, clock issues, RAM design flaws, CPU/RAM/Motherboard
integrated logic defects, DRAM cell amplification errors, cosmic rays
How many bitflips can be corrected and detected? - CORRECT ANSWER - Error
Correcting Code Memory (ECC memory) can correct 1 bitflip / detect 2 bitflips by using TMR
or Hamming Code
Cell types of NAND flash storage in SSDs - CORRECT ANSWER - SSDs consist of
NAND cells with a limited lifetime, following types exist:
SLC (single-level cells): 10'000 - 100'000 write/erase cycles [1 bit per cell]
UPDATED ACTUAL Exam Questions and
CORRECT Answers
Why do we need distributed systems? - CORRECT ANSWER - Scaling (if one machine is
not enough), Location (to move closer to the user), Fault-Tolerance (HW will eventually fail)
Vertical Scaling - CORRECT ANSWER - more memory, faster CPU (scale up)
Horizontal Scaling - CORRECT ANSWER - more machines (scale out)
Scaling for Machine Learning - CORRECT ANSWER - Current Trend: Scale horizontally,
Building models, fine Tuning → faster with multiple cards (1 User = 1 Card), storing data →
scalable storage, ML with vertical scaling is not possible
Pros / Cons Horizontal Scaling - CORRECT ANSWER - + Lower cost with massive scale
+ Easier to add fault-tolerance
+ Higher availability
− Adaption of software required
− More complex system, more components involved
Pros / Cons Vertical Scaling - CORRECT ANSWER - + Lower cost with small scale
+ No adaption of software required
+ Less complexity
− HW limits for scaling
− Risk of HW failure causing outage
− More difficult to add fault-tolerance
, Moore's Law - CORRECT ANSWER - Nr. of transistors on microchips doubles every two
years (other predictions, doubling chip performance every 18 month)
Nielsen's Law - CORRECT ANSWER - a high-end user's connection speed grows by 50%
per year, bandwidth grows slower than computer power → telecom companies are conservative,
user are reluctant to spend much money on bandwidth, the user base is getting broader
Kryder's Law - CORRECT ANSWER - disk density doubling every 13 months
Latency - CORRECT ANSWER - time for a signal to travel from source to destination and
back (RTT - round-trip time)
Why should one place a service closer to a user? - CORRECT ANSWER - Reduced
latency, can increase bandwidth and throughput, can improve reliability and availability,
drawback: coordination of data replication and caching
CDN - Content Delivery Network - CORRECT ANSWER - also: Content distribution
network, distributed databases, edge computing, placing images, sites, scripts, close to the
respective user (physically)
Why do random bit flips in memory occur? - CORRECT ANSWER - Bad Pin
connections, incorrect RAM timings, clock issues, RAM design flaws, CPU/RAM/Motherboard
integrated logic defects, DRAM cell amplification errors, cosmic rays
How many bitflips can be corrected and detected? - CORRECT ANSWER - Error
Correcting Code Memory (ECC memory) can correct 1 bitflip / detect 2 bitflips by using TMR
or Hamming Code
Cell types of NAND flash storage in SSDs - CORRECT ANSWER - SSDs consist of
NAND cells with a limited lifetime, following types exist:
SLC (single-level cells): 10'000 - 100'000 write/erase cycles [1 bit per cell]
