Leaving aside the political firestorm, is the book good?
By literary standards, Murtad is uneven. The prose is functional. The plot occasionally leans on coincidences. It is not Crime and Punishment.
But that misses the point.
Murtad is not a great novel. It is a vital novel. It is a pressure valve. In a society where you cannot openly say "I am an atheist" without risking your job, your family, or your freedom, you write a pulp novel about a gangster who loses his faith. murtad fixi pdf
The book’s legacy is not in its sentences, but in its existence. It proved that the Malaysian publishing scene was not a monolith. It proved that the young, urban Malay demographic is far more complex than the politicians believe.
The policy engine formulates the injection decision as a Markov Decision Process (MDP):
[ r = \alpha \cdot \textCoverage - \beta \cdot \textSLA_Violation - \gamma \cdot \textOverhead ] Leaving aside the political firestorm, is the book good
where Coverage is the proportion of unique fault‑type‑intensity pairs exercised, SLA_Violation measures latency breaches, and Overhead reflects CPU usage of the FI agent.
We employ Proximal Policy Optimization (PPO) with a lightweight neural network (2 hidden layers, 64 units each) running on the orchestrator. The policy updates every 30 seconds using a mini‑batch of the latest 500 telemetry samples.
| Metric | MURTAD‑FIXI | Chaos Monkey | NetEm (static) | FI‑Tool | |--------|------------|--------------|----------------|---------| | Fault Coverage (unique combos) | 342 | 84 | 71 | 115 | | Avg. Overhead (CPU %) | 4.7 % | 2.1 % | 1.8 % | 6.3 % | | Bugs Found (per hour) | 17 | 5 | 3 | 8 | | SLA Compliance | 96 % | 92 % | 94 % | 89 % | [ r = \alpha \cdot \textCoverage - \beta
Figure 1 (not shown) illustrates the reward curve over a 2‑hour run, highlighting rapid convergence after ~30 minutes.
Key observations
| Approach | Target Platform | Fault Types | Intrusiveness | Adaptivity | |----------|----------------|-------------|---------------|------------| | Chaos Monkey (Netflix) | Cloud VMs/Containers | Process/Pod termination | Low (API‑based) | None | | FI‑Tool (Intel) | x86 Simulators | CPU, cache, memory | High (simulation) | Static | | GEM5 | Architectural simulators | Micro‑architectural | Very high | None | | Eclipse DEFT | Embedded RTOS | ISR delays, memory corruption | Medium (source instrumentation) | None | | SAPHIRE | Distributed systems | Network partitions | Low (iptables) | Static (pre‑defined) | | MURTAD‑FIXI | Heterogeneous edge nodes | Software, network, hardware, power | Very low (eBPF & containers) | RL‑based, dynamic |
Recent works such as EdgeChaos and Fidelity have introduced network‑centric fault injection for edge clusters, but they still lack hardware‑level fault models and online learning to adapt to runtime metrics. MURTAD‑FIXI fills this void by unifying all fault domains under a common policy engine.