Distributed Game Engine Paper Reading
DHR: Distributed Hybrid Rendering for Metaverse Experiences DHR: Distributed Hybrid Rendering for Metaverse Experiences The client performs rasterization locally, while the server computes ray...
HYJun Bell
Distributed Path Tracer
Motivation In traditional cloud gaming architecture, a game instance is running in the cloud. The thin client sends users’ input to cloud gaming server. The server does the game logics computation...
Software Grand Exposure - SGX Cache Attacks Are Practical
Introduction Intel Software Guard Extensions (SGX) enables secure execution of sensitive application code (enclaves) by isolating it from untrusted system software. SGX provides protections i...
ZeRO-Offload - Democratizing Billion-Scale Model Training
Large model training–Scale out Use aggregate memory of multiple GPUs to satisfy the memory requirement. Model parallelism partitions the model vertically and distributes the model partitions ...
Distributed Machine Learning Paper Reading
Large Scale Distributed Deep Networks Large Scale Distributed Deep Networks Only thick lines (interconnections on different machines) need to communicate, and even if there are multiple edge...
Eculid - A DDoS Detection Framwork On Programmable Switch
Eculid Euclid is a fully in-network DDoS detection system. Offloads both the detection and mitigation algorithms to the data plane. Uses Shannon entropy value of IP addresses for detection....
Virtualization
Types of Virtualization Type 1 (Baremetal) The Virtual Machine Monitor (VMM) manages all hardware resources and supports the execution of Virtual Machines (VMs). The hypervisor ...
Tenplex - Dynamic Parallelism for Deep Learning using Parallelizable Tensor Collections
Need for dynamic resource changes Elasticity: Elastically increase and decrease the allocated GPUs for a job based on the available resources. Redeployment: Reallocate jobs to a new set of GP...