ashwanth kumar @_ashwanthkumar software engineer suuchi - library of distributed systems primitives 

blank yep, it’s intentional 

from simple to unmanageable beasts how we scale systems? an opinionated view 

Service 

Service 

Service Service 

Service Service Service 

Service Service Service 

Service Service Service 

Service Service Service 

rise of KV stores distributed, replicated, fault-tolerant (optionally sorted) 2006 BigTable from Google 2007 Dynamo from Amazon 2009 VoldemortDB from LinkedIn Cassandra from facebook 2008 

Distributed (NoSQL) Datastores Service Service Service 

Distributed (NoSQL) Datastores Service Service Service Network Latency 

data locality for low latency / data intensive applications 

Service Service Service Co-locate data to improve performance 

Sharded and replicated to improve throughput Service Service Service 

Service Service Service Deal with complex distributed system problems at the application layer 

1.1.1.1 1.1.1.2 1.1.1.3 A stats.service.ix 1.1.1.1 1.1.1.2 1.1.1.3 DNS based Load Balancing Distributed stats aggregation system @indix Count(“a”, 1L) Unique(“a”, 1L) Count(“c”, 1L) Unique(“a”, 1L) Count(“b”, 1L) monoid.plus monoid.plus monoid.plus 

पांग ப Communication key=”foo” key=”bar” key=”baz” Request Routing Sync / Async Replication Replication Data Sharding Cluster Membership distributed system problems 

suuchi github.com/ashwanthkumar/suuchi library of distributed systems primitives 

+ HandleOrForward - Broadcast - ScatterGather primitives - communication uses http/2 with streaming 

+ Consistent Hash Ring - Your own sharding technique? primitives - sharding/routing Consistent hashing and random trees: Distributed caching protocols for relieving hot spots on the World Wide Web node 2 node 1 node 3 node 4 

primitives - membership static dynamic fault tolerance in case of node/process failure scaling up/down needs downtime of the system 

primitives - replication sync async provides very high availability for write systems at the cost of eventual consistency every request is successful only if all the replicas succeeded 

+ KeyValue + RocksDB - Your Own Abstraction? primitives - storage embedded KV store from FB for server workloads 

- define a gRPC service using proto2 (or proto3) - generate the stubs in java / scala - implement the services - connect them together using Suuchi - Server abstraction getting started 

let’s see some code see reference for actual links 

- HTML Archive System - Handles 1000+ tps - write heavy system - Stores ~120TB of url and timestamp indexed HTML pages - Stats (as Monoids) Storage System* - All we want is approximate aggregates real-time - Real-time scheduler for our crawlers* - Finds out which of the 20 urls to crawl now out of 3+ billion urls - Helps crawler crawl 20+ million urls everyday suuchi @indix 

idea behind suuchi membership, request routing / sharding 2011 Gizzard from Twitter 2016 Suuchi 2016 Slicer from Google 2015 RingPop from Uber 

questions? references available at github.com/ashwanthkumar/suuchi-ds-primitives 

more on consistent hash ring 

primitives - request routing Consistent hashing and random trees: Distributed caching protocols for relieving hot spots on the World Wide Web node 2 node 1 node 3 node 4 

primitives - request routing Consistent hashing and random trees: Distributed caching protocols for relieving hot spots on the World Wide Web node 2 node 1 node 3 node 4 

primitives - request routing Consistent hashing and random trees: Distributed caching protocols for relieving hot spots on the World Wide Web node 2 node 1 node 3 node 4 

primitives - request routing Consistent hashing and random trees: Distributed caching protocols for relieving hot spots on the World Wide Web node 2 node 3 node 4 

- Peer to Peer system - no single point of contact - Each node handles or forwards requests transparently - Uses pluggable partitioner scheme - Can be customized as weighted distribution / Rendezvous Hash etc. primitives - request routing