Author: lechuck park

AI Infrastructure Architect & Technical Visualizer "Complex Systems, Simplified. I translate massive AI infrastructure into visual intelligence." I love to learn computer tech and help people by the digital.

Page(Memory) Replacement with AI

Posted on by lechuck park

With Claude
This image illustrates a Page (Memory) Replacement system using AI. Let me break down the key components:

  1. Top Structure:
  • Paging (Legacy & current): Basic paging system structure
  • Logical Memory: Organized in 4KB units, maximum 64-bit sizing (2^64 Bytes)
  • Physical Memory: Limited to the actual installed memory size
  1. Memory Allocation:
  • Shows Alloc() and Dealloc() functions
  • When no more allocation is possible, there’s a question about deallocation strategy:
    • FIFO (First In First Out): Deallocate the oldest allocated memory first
    • LRU (Least Recently Used): Deallocate the oldest used memory first
  1. AI-based Page Replacement Process:
  • Data Collection: Gathers information about page access frequency, time intervals, and memory usage patterns
  • Feature Extraction: Analyzes page access time, access frequency, process ID, memory region, etc.
  • Model Training: Aims to predict the likelihood of specific pages being accessed in the future
  • Page Replacement Decision: Pages with a low likelihood of future access are prioritized for swapping
  • Real-Time Application & Evaluation: Applies the model in real-time to perform page replacement and evaluate system performance

This system integrates traditional page replacement algorithms with AI technology to achieve more efficient memory management. The use of AI helps in making more intelligent decisions about which pages to keep in memory and which to swap out, based on learned patterns and predictions.

Analysis Evolutions and ..

Posted on by lechuck park

With Claude
this image that shows the evolution of data analysis and its characteristics at each stage:

Analysis Evolution:

  1. 1-D (One Dimensional): Current Status analysis
  2. Time Series: Analysis of changes over time
  3. n-D Statistics: Multi-dimensional correlation analysis
  4. ML/DL (Machine Learning/Deep Learning): Huge-dimensional analysis including exceptions

Bottom Indicators’ Changes:

  1. Data/Computing/Complexity:
  • Marked as “Up and Up” and increases “Dramatically” towards the right
  1. Accuracy:
  • Left: “100% with no other external conditions”
  • Right: “not 100%, up to 99.99% from all data”
  1. Comprehensibility:
  • Left: “Understandable/Explainable”
  • Right: “Unexplainable”
  1. Actionability:
  • Left: “Easy to Action”
  • Right: “Difficult to Action require EXP” (requires expertise)

This diagram illustrates the trade-offs in the evolution of data analysis. As analysis methods progress from simple one-dimensional analysis to complex ML/DL, while the sophistication and complexity of analysis increase, there’s a decrease in comprehensibility and ease of implementation. It shows how more advanced analysis techniques, while powerful, require greater expertise and may be less transparent in their decision-making processes.

The progression also demonstrates how modern analysis methods can handle increasingly complex data but at the cost of reduced explainability and the need for specialized knowledge to implement them effectively.

TIMELY

Posted on by lechuck park

With Claude
TIMELY (Transport Informed by MEasurement of LatencY)

  1. System Architecture
  • Cloud/Data Center to External Network Connection
  • TIMELY Module Process at Kernel Level
  • Bidirectional Operation Support
  • TCP Protocol Based
  1. RTT-based Traffic Control Components
  • RTT Monitoring
    • 5-tuple monitoring (Src/Dst IP, Src/Dst Port, Protocol)
    • Real-time latency measurement
  • Congestion Detection
    • Network congestion detection through RTT increases
  • Congestion Window Adjustment
    • Control of send buffer size
  • MSS-based Adjustments
    • Congestion window adjustments in MSS units
  1. Related RTT-based Technologies
  • TCP BBR
  • TCP Vegas
  • CUBIC TCP
  1. Advantages of RTT-based Control
  • Proactive congestion detection before packet loss
  • Real-time network state awareness
  • Efficient buffer management
  • Lower latency in data transmission
  • Effective bandwidth utilization
  • Better performance in high-speed networks
  1. Disadvantages of RTT-based Control
  • RTT measurement accuracy dependency
  • Complex implementation at kernel level
  • Potential overhead in RTT monitoring
  • Need for continuous RTT measurement
  • Sensitivity to network jitter
  • May require adjustments for different network environments

The TIMELY system demonstrates an efficient approach to network congestion control using RTT measurements, particularly suitable for cloud and data center environments where latency and efficient data transmission are critical. The system’s kernel-level implementation and MSS-based adjustments provide fine-grained control over network traffic, though success heavily depends on accurate RTT measurements and proper environment calibration.

Deepseek

Posted on by lechuck park

With Claude
The evolution pipeline of the Deepseek model consists of three major stages:

Stage 1: V3-Base → R1-Zero

  • Direct application of Reinforcement Learning (RL)
  • Proceeds without Supervised Fine-tuning (SFT)
  • Adopts learning approach toward exact reward
  • Performs basic data classification tasks

Stage 2: R1-Zero → R1

  • Utilizes cold-start data for learning
  • Implements multi-stage training pipeline
  • Conducts foundational learning with initial data
  • Applies systematic multi-stage learning process

Stage 3: R1 → R1-Distill-(XXX)

  • Model optimization through knowledge distillation
  • Smaller models achieve excellent performance through SFT alone
  • Continuous model tuning through evaluations
  • Performance enhancement through learning with other models

This pipeline demonstrates a comprehensive approach to model development, incorporating various advanced AI training techniques and methodologies to achieve optimal performance at each stage.

Von Neumann architecture / Neuromorphic computing

Posted on by lechuck park

With Claude
This image illustrates the comparison between Von Neumann architecture and Neuromorphic computing.

The upper section shows the traditional Von Neumann architecture:

  1. It has a CPU (Operator) that processes basic operations (+, -, ×, =) sequentially
  2. Data is brought from memory (“Bring all from memory”) and processed in sequence
  3. All operations are performed sequentially (“Sequential of operator”)

The lower section demonstrates Neuromorphic computing:

  1. It shows a neural network structure where multiple nodes are interconnected
  2. Each connection has different weights (“Different Weight”) and performs simple operations (“Simple Operate”)
  3. All operations are processed in parallel (“Parallel Works”)

Key differences between these architectures:

  • Von Neumann architecture: Sequential processing, centralized computation
  • Neuromorphic computing: Parallel processing, distributed computation, design inspired by the human brain’s structure

The main advantage of Neuromorphic computing is that it provides a more efficient architecture for artificial intelligence and machine learning tasks by mimicking the biological neural networks found in nature. This parallel processing approach can handle complex computational tasks more efficiently than traditional sequential processing in certain applications.

The image effectively contrasts how data flows and is processed in these two distinct computing paradigms – the linear, sequential nature of Von Neumann versus the parallel, interconnected nature of Neuromorphic computing.