Chapter 2 - Compression Techniques “I have made this longer than usual because I have not had time to make it shorter.” Blaise Pascal In the last chapter, we discussed a few ideas to improve the deep deep learning efficiency. Now, we will elaborate on one of those ideas, the compression techniques. Compression techniques aim to reduce the model footprint (size, latency, memory etc.). We can reduce the chapter, we introduce Quantization, a model compression technique that addresses both these issues. We’ll start with a gentle introduction to the idea of compression. Details of quantization and its applications

== 3 9 . 4/ String Compression Lets make a compressed string table https://github.com/AshleyRoll/squeeze map from enum Key to Compressed String Hu�man Coding for compression Output struct: Mapping choose an arbitrary amount of work they will allow in constexpr context Complex processing like compression will hit the limits Had to make more complex implementation to cache bit streams rather than walk

Advanced Compression Techniques “The problem is that we attempt to solve the simplest questions cleverly, thereby rendering them unusually complex. One should seek the simple solution.” — Anton Pavlovich Pavlovich Chekhov In this chapter, we will discuss two advanced compression techniques. By ‘advanced’ we mean that these techniques are slightly more involved than quantization (as discussed in the second of our models. Did we get you excited yet? Let’s learn about these techniques together! Model Compression Using Sparsity Sparsity or Pruning refers to the technique of removing (pruning) weights during

50 About Validators 51 About Comparators 51 About Column Family Compression 52 When to Use Compression 52 Configuring Compression on a Column Family 52 About Indexes in Cassandra 52 About Primary compaction_strategy 80 compaction_strategy_options 80 comparator 81 compare_subcolumns_with 81 compression_options 81 default_validation_class 81 gc_grace_seconds 81 key_cache_save_period_in_seconds of columns. About Validators 51 About Column Family Compression Data compression can be configured on a per-column family basis. Compression maximizes the storage capacity of your Cassandra nodes by

Core Features 209 INTERVAL Data Type Handling 209 Additional PostgreSQL Features 210 Zstandard Compression Algorithm 210 Relaxed Rules for Specifying Table Distribution Columns 210 Resource Groups Features 0 Data Model 315 Heap vs. Append-Optimized Storage 315 Row vs. Column Oriented Storage 315 Compression 316 Distributions 316 Resource Queue Memory Management 316 Partitioning 318 Indexes 318 Resource Storage Model 328 Heap Storage or Append-Optimized Storage 329 Row or Column Orientation 329 Compression 330 Distributions 330 Local (Co-located) Joins 331 Data Skew 331 Processing Skew 332 Partitioning

Core Features 217 INTERVAL Data Type Handling 217 Additional PostgreSQL Features 218 Zstandard Compression Algorithm 218 Relaxed Rules for Specifying Table Distribution Columns 218 VMware Greenplum v6 0 Data Model 324 Heap vs. Append-Optimized Storage 324 Row vs. Column Oriented Storage 324 Compression 325 Distributions 325 Resource Queue Memory Management 325 Partitioning 327 Indexes 327 Resource Storage Model 337 Heap Storage or Append-Optimized Storage 338 Row or Column Orientation 338 Compression 339 Distributions 339 Local (Co-located) Joins 340 Data Skew 340 Processing Skew 341 Partitioning

application itself. Baseline Compression settings are the main factor affecting the baseline amount of memory used by each connection. By default websockets maximizes compression rate at the expense of memory memory usage. If memory usage is an issue, lowering compression settings can help: • Context Takeover is necessary to get good performance for almost all applications. It should remain enabled. • Window Window Bits is a trade-off between memory usage and compression rate. It defaults to 15 and can be lowered. The default value isn’t optimal for small, repetitive messages which are typical of WebSocket servers

Core Features 269 INTERVAL Data Type Handling 269 Additional PostgreSQL Features 270 Zstandard Compression Algorithm 271 Relaxed Rules for Specifying Table Distribution Columns 271 Resource Groups Features 567 To create a column-oriented table 568 Using Compression (Append-Optimized Tables Only) 569 To create a compressed table 570 Checking the Compression and Distribution of an Append-Optimized Table 570 by Broadcom 27 Adding Column-level Compression 572 Default Compression Values 573 Precedence of Compression Settings 573 Optimal Location for Column Compression Settings 574 Storage Directives Examples

0 Data Model 248 Heap vs. Append-Optimized Storage 248 Row vs. Column Oriented Storage 248 Compression 249 Distributions 249 Resource Queue Memory Management 249 VMware Greenplum v6.17 Documentation Storage Model 261 Heap Storage or Append-Optimized Storage 262 Row or Column Orientation 262 Compression 263 Distributions 263 Local (Co-located) Joins 264 Data Skew 264 Processing Skew 265 VMware Configuration Parameter Categories 442 Configuration Parameter Categories 0 Enabling Compression 442 Enabling Compression 0 Configuring Proxies for the Greenplum Interconnect 443 Configuring Proxies for

Core Features 214 INTERVAL Data Type Handling 214 Additional PostgreSQL Features 215 Zstandard Compression Algorithm 215 Relaxed Rules for Specifying Table Distribution Columns 215 Resource Groups Features vs. Column Oriented Storage 319 VMware Tanzu Greenplum v6.20 Documentation VMware, Inc. 14 Compression 320 Distributions 320 Resource Queue Memory Management 320 Partitioning 322 Indexes 322 Resource Storage Model 332 Heap Storage or Append-Optimized Storage 333 Row or Column Orientation 333 Compression 334 Distributions 334 Local (Co-located) Joins 335 Data Skew 335 Processing Skew 336 VMware