Research Session 22 Modern Hardware 1
Table of Contents (powered by https://videoken.com)
0:02:36 [Paper: LATTE: A Native Table Engine on NVMe Storage]
0:03:20 Background
0:04:45 Motivation
0:07:06 Design
0:07:25 Parallel Queues Scheduling
0:08:58 Binding of I/O Quesues and CPU Cores
0:10:53 Undo Logging on Heterogeneous Storage
0:12:18 Evaluation
0:15:34 Conclusion
0:16:21 Thank you Q&A
0:20:30 [Paper: Doubleheader Logging: Eliminating Journal Write Overhead of Mobile DBMS]
0:20:56 Motivation: Journaling of Journal
0:21:30 If WAL is Used
0:22:31 Can we eliminate Journaling of Journal?
0:22:39 Motivation of Doubleheader Logging
0:23:17 Background: Slotted Page Structure
0:24:22 Doubleheader Logging
0:26:31 DHL vs. WAL
0:30:12 Counting Commit
0:30:34 Experimental Setup
0:31:13 Performance Breakdown
0:32:07 EXT4 Block Trace
0:33:30 Real Workload
0:34:01 Conclusion
0:38:03 [Paper: GSI: GPU-friendly Subgraph Isomorphism]
0:38:29 Graphs are Everywhere
0:38:38 Subgraph Match
0:38:52 Running Example
0:39:10 Major Challenges
0:39:33 GPU Architecture
0:40:17 Edge-based Join
0:40:53 Edge-based Join in GPU [1,2]
0:41:46 Vertex-based Join
0:42:53 Our Contributions
0:43:04 Encoding-based Pruning
0:44:10 Two primitive operations
0:44:45 PCSR: A new data structure
0:48:43 GSI: Parallel Join Algorithm
0:50:39 Some Optimizations
0:50:45 GPU-Friendly Set Operations
0:51:01 Addressing Skew Degree Distribution
0:51:33 Duplicate Removal
0:51:49 Experiments
0:52:34 Conclusions
0:52:54 References
0:52:57 Thanks
0:56:52 [Paper: FGPA-based Compaction Engine for Accelerating LSM-tree Key-Value Stores]
0:57:15 LSM-tree based KV Stores* are Popular Now!
0:57:52 Background
0:59:02 Write Pause
1:00:21 Database Acceleration with FPGAs*
1:01:17 Challenges for Compaction Offloading
1:02:04 Implementation
1:02:08 SSTable Format
1:02:44 System Overview
1:03:20 Basic Compaction Engine
1:04:10 Optimization 1
1:04:45 Optimization 2
1:05:15 Optimization 3
1:06:35 Software Integration
1:07:30 Evaluation
1:07:35 Experimental Setup
1:08:22 Compaction Speed
1:08:55 Micro Benchmark
1:09:52 YCSB Benchmark
1:10:28 Thanks!
1:13:41 [Paper: Getting Swole: Generating Access-Aware Code with Predicate Pullups]
1:14:55 Tweet
1:15:11 What is code generation?
1:16:57 What are predicate pullups?
1:18:03 What does "Swole" mean?
1:18:28 Existing Strategies
1:20:05 Strategy #3: ROF
1:20:07 Existing Strategies
1:21:22 Access Patterns
1:22:24 Swole
1:23:19 Swole Techniques
1:23:25 Value Masking
1:26:44 Selected Related Work
1:27:28 Evaluation
1:28:16 Please check out the paper!
1:28:18 Main Takeaways