EVCache: Lowering Costs for a Low Latency Cache with RocksDB Scott Mansfield Vu Nguyen EVCache
90 seconds
What do caches touch? Signing up* Logging in Choosing a profile Picking liked videos Personalization* Loading home page* Scrolling home page* A/B tests Video image selection * multiple caches involved Searching* Viewing title details Playing a title* Subtitle / language prefs Rating a title My List Video history* UI strings Video production*
Home Page Request
Ephemeral Volatile Cache Key-Value store optimized for AWS and tuned for Netflix use cases
What is EVCache? Distributed, sharded, replicated key-value store Tunable in-region and global replication Based on Memcached Resilient to failure Topology aware Linearly scalable Seamless deployments
Why Optimize for AWS Instances disappear Zones fail Regions become unstable Network is lossy Customer requests bounce between regions Failures happen and we test all the time
EVCache Use @ Netflix Hundreds of terabytes of data Trillions of ops / day Tens of billions of items stored Tens of millions of ops / sec Millions of replications / sec Thousands of servers Hundreds of instances per cluster Hundreds of microservice clients Tens of distinct clusters 3 regions 4 engineers
Architecture Application Server Client Library Client Memcached EVCar Eureka (Service Discovery)
Architecture us-west-2a Client us-west-2b Client us-west-2c Client
Reading (get) us-west-2a us-west-2b Client Primary Secondary us-west-2c
Writing (set, delete, add, etc.) us-west-2a Client us-west-2b Client us-west-2c Client
Use Case: Lookaside Cache Application Client Library Client Ribbon Client S Data Flow C S S C C S C
Use Case: Transient Data Store Time Application Application Client Library Application Client Library Client Client Library Client Client
Use Case: Primary Store Online Application Client Library Client Online Services Offline Services Offline / Nearline Precomputes for Recommendations Data Flow
Use Case: Versioned Primary Store Online Application Archaius (Dynamic Properties) Client Library Client Online Services Offline Services Offline Compute Data Flow Control System (Valhalla)
Use Case: High Volume && High Availability Application Client Library Optional In-memory Remote Compute & Publish on schedule Data Flow Ribbon Client S C S C S S C C
Pipeline of Personalization Online 1 Online 2 Online Services Offline Services Compute A Compute B Data Flow Compute D Compute C Compute E
Additional Features Kafka Global data replication Consistency metrics Key Iteration Cache warming Lost instance recovery Backup (and restore)
Additional Features (Kafka) Global data replication Consistency metrics
Cross-Region Replication Region A APP Region B 7 read 1 mutate APP a t da t ge t 4 r se fo 6 mutate 2 send metadata Repl Proxy Repl Proxy sg s tp 5 Kafka 3 poll msg Repl Relay ht nd se m Repl Relay Kafka
Additional Features (Key Iteration) Cache warming Lost instance recovery Backup (and restore)
Cache Warming Application Client Library Client Control Data Flow Control Flow Metadata Flow S3 Cache Warmer (Spark)
Moneta Next-generation EVCache server
Moneta Moneta: The Goddess of Memory Juno Moneta: The Protectress of Funds for Juno Evolution of the EVCache server Cost optimization EVCache on SSD Ongoing lower EVCache cost per stream Takes advantage of global request patterns
Old Server Stock Memcached and EVCar (sidecar) All data stored in RAM in Memcached Expensive with global expansion / N+1 architecture Memcached EVCar external
Optimization Global data means many copies Access patterns are heavily region-oriented In one region: Hot data is used often Cold data is almost never touched Keep hot data in RAM, cold data on SSD Size RAM for working set, SSD for overall dataset
New Server Adds Rend and Mnemonic Still looks like Memcached Unlocks cost-efficient storage & server-side intelligence
Rend go get github.com/netflix/rend
Rend High-performance Memcached proxy & server Written in Go Powerful concurrency primitives Productive and fast Manages the L1/L2 relationship Tens of thousands of connections
Rend Modular set of libraries and an example main() Manages connections, request orchestration, and communication Low-overhead metrics library Connection Management Multiple orchestrators Server Loop Protocol Parallel locking for data integrity Request Orchestration Efficient connection pool Backend Handlers M E T R I C S
Mnemonic
Mnemonic Manages data storage on SSD Uses Rend server libraries Handles Memcached protocol Maps Memcached ops to RocksDB ops Rend Server Core Lib (Go) Mnemonic Op Handler (Go) Mnemonic Core (C++) RocksDB (C++)
Why RocksDB? Fast at medium to high write load Disk--write load higher than read load (because of Memcached) Predictable RAM Usage Record A Record B memtable memtable SST SST memtable SST SST: Static Sorted Table...
How we use RocksDB No Level Compaction Generated too much traffic to SSD High and unpredictable read latencies No Block Cache Rely on Local Memcached No Compression
How we use RocksDB FIFO Compaction SST s ordered by time Oldest SST deleted when full Reads access every SST until record found
How we use RocksDB Full File Bloom Filters Full Filter reduces unnecessary SSD reads Bloom Filters and Indices pinned in memory Minimize SSD access per request
How we use RocksDB Records sharded across multiple RocksDB per node Reduces number of files checked to decrease latency Mnemonic Core Key: ABC Key: XYZ... R R R R
Region-Locality Optimizations Replication and Batch updates only RocksDB* Keeps Region-Local and hot data in memory Separate Network Port for off-line requests Memcached data replaced
FIFO Limitations FIFO compaction not suitable for all use cases Very frequently updated records may push out valid records Expired Records still exist Requires Larger Bloom Filters SST SST SST Record A1 Record A1 Record C Record A2 Record A2 Record D Record B1 Record B1 Record E Record A3 Record A3 Record F Record B2 Record B2 Record G Record B3 Record B3 Record H time
AWS Instance Type i2.xlarge 4 vcpu 30 GB RAM 800 GB SSD 32K IOPS (4KB Pages) ~130MB/sec
Moneta Perf Benchmark (High-Vol Online Requests)
Moneta Perf Benchmark (cont)
Moneta Perf Benchmark (cont)
Moneta Perf Benchmark (cont)
Moneta Performance in Production (Batch Systems) Request get 95%, 99% = 729 μs, 938 μs L1 get 95%, 99% = 153 μs, 191 μs L2 get 95%, 99% = 1005 μs, 1713 μs ~20 KB Records ~99% Overall Hit Rate ~90% L1 Hit Rate
Moneta Performance in Prod (High Vol-Online Req) Request get 95%, 99% = 174 μs, 588 μs L1 get 95%, 99% = 145 μs, 190 μs L2 get 95%, 99% = 770 μs, 1330 μs ~1 KB Records ~98% Overall Hit Rate ~97% L1 Hit Rate
Moneta Performance in Prod (High Vol-Online Req) Get Percentiles: 50th: 102 μs (101 μs) 75th: 120 μs (115 μs) 90th: 146 μs (137 μs) 95th: 174 μs (166 μs) 99th: 588 μs (427 μs) 99.5th: 733 μs (568 μs) 99.9th: 1.39 ms (979 μs) Latencies: peak (trough) Set Percentiles: 50th: 97.2 μs (87.2 μs) 75th: 107 μs (101 μs) 90th: 125 μs (115 μs) 95th: 138 μs (126 μs) 99th: 177 μs (152 μs) 99.5th: 208 μs (169 μs) 99.9th: 1.19 ms (318 μs)
70% Reduction in cost*
Challenges/Concerns Less Visibility Unclear of Overall Data Size because of duplicates and expired records Restrict Unique Data Set to ½ of Max for Precompute Batch Data Lower Max Throughput than Memcached-based Server Higher CPU usage Planning must be better so we can handle unusually high request spikes
Current/Future Work Investigate Blob Storage feature Less Data read/write from SSD during Level Compaction Lower Latency, Higher Throughput Better View of Total Data Size Purging Expired SST s earlier Useful in short TTL use cases May purge 60%+ SST earlier than FIFO Compaction Reduce Worst Case Latency Better Visibility of Overall Data Size Inexpensive Deduping for Batch Data
Open Source https://github.com/netflix/evcache https://github.com/netflix/rend
Thank You smansfield@netflix.com (@sgmansfield) nguyenvu@netflix.com techblog.netflix.com
Failure Resilience in Client Operation Fast Failure Tunable Retries Operation Queues Tunable Latch for Mutations Async Replication through Kafka
Consistency Metrics Region A APP 1 mutate Client Dashboards 2 send metadata 4 pull data Kafka 3 poll msg Consistency Checker Atlas (Metrics Backend) 5 report
Lost Instance Recovery Application Client Library Client Zone A Zone B Control Data Flow Partial Data Flow Control Flow Metadata Flow S3 Cache Warmer (Spark)
Backup (and Restore) Application Client Library Client Control Data Flow Control Flow S3 Cache Warmer (Spark)
Moneta in Production Serving all of our personalization data Rend runs with two ports: One for standard users (read heavy or active data management) Another for async and batch users: Replication and Precompute Maintains working set in RAM Optimized for precomputes Smartly replaces data in L1 Std Memcached (RAM) Batch Mnemonic EVCar external internal (SSD)
Rend batching backend