Are We Ready for Serverless
Subbu Allamaraju
Container World | Feb 27, 2018 | Santa Clara, CA
@sallamar
https://www.subbu.org
More than 50% of
compute may be
stateless
Copyright 2018 Expedia Inc. 2
Observation
Hypothesis
Copyright 2018 Expedia Inc. 3
Most of stateless is a
candidate for some
kind of serverless
Copyright 2018 Expedia Inc. 4
Hypothesis
Most of state is
moving to managed
cloud services
Capacity demand
Desired capacity
Actual allocated capacity + additional DR capacity?
A dream from 2011
Copyright 2018 Expedia Inc. 5
Reality in 2018
Copyright 2018 Expedia Inc. 6
A few billion lambda invocations per month
What about state?
S3, Glacier, Aurora,
Elasticache, RDS,
Redshift, DynamoDB,
Neptune, Athena,
ElasticSearch Service,
Quicksight, EMR,
Kinesis, Data Pipeline,
CloudSearch, Redshift,
SQS, SNS, MQ, …
BigQuery, Cloud
Dataflow, Cloud
Dataproc, Cloud
Datalab, Cloud
Dataprep, Cloud
Pub/Sub, Data Studio,
Cloud SQL, Cloud
Bigtable, Cloud
Spanner, Cloud
Datastore, …
Queue Storage, Blob
Storage, Notification
Hubs, Search, SQL
Database, SQL Data
Warehouse, SQL Server
Stretch Database,
Cosmos DB, Redis
Cache, Table Storage,
HDInsight, SQL Data
Warehouse, Event Hubs,
…
Copyright 2018 Expedia Inc. 7
Container ecosystem may
or may not play a role
Copyright 2018 Expedia Inc. 8
Hypothesis
Are we really ready?
Copyright 2018 Expedia Inc. 9
Not suitable for long-running
appsExpensive
Lock-in
Slow on cold start
No local storage
More complex call patterns
Unknown security risks
No standards
Observability is hard
?
Copyright 2018 Expedia Inc. 10
Habits and inertia
Stateless isn’t always
stateless
Ephemeral isn’t
always ephemeral
Often pet like
?
Copyright 2018 Expedia Inc. 11
Fear of the ecosystem
Lock-in worries
Undifferentiated heavy
lifting vs Value
generation?
?
Copyright 2018 Expedia Inc. 12
Maturity
Feature gaps
Pricing
Tomorrow != Today
Tooling?
Copyright 2018 Expedia Inc. 13
Embrace the paradigm
Learn
Get ready
Copyright 2018 Expedia Inc. 14

Are We Ready for Serverless

  • 1.
    Are We Readyfor Serverless Subbu Allamaraju Container World | Feb 27, 2018 | Santa Clara, CA @sallamar https://www.subbu.org
  • 2.
    More than 50%of compute may be stateless Copyright 2018 Expedia Inc. 2 Observation
  • 3.
    Hypothesis Copyright 2018 ExpediaInc. 3 Most of stateless is a candidate for some kind of serverless
  • 4.
    Copyright 2018 ExpediaInc. 4 Hypothesis Most of state is moving to managed cloud services
  • 5.
    Capacity demand Desired capacity Actualallocated capacity + additional DR capacity? A dream from 2011 Copyright 2018 Expedia Inc. 5
  • 6.
    Reality in 2018 Copyright2018 Expedia Inc. 6 A few billion lambda invocations per month
  • 7.
    What about state? S3,Glacier, Aurora, Elasticache, RDS, Redshift, DynamoDB, Neptune, Athena, ElasticSearch Service, Quicksight, EMR, Kinesis, Data Pipeline, CloudSearch, Redshift, SQS, SNS, MQ, … BigQuery, Cloud Dataflow, Cloud Dataproc, Cloud Datalab, Cloud Dataprep, Cloud Pub/Sub, Data Studio, Cloud SQL, Cloud Bigtable, Cloud Spanner, Cloud Datastore, … Queue Storage, Blob Storage, Notification Hubs, Search, SQL Database, SQL Data Warehouse, SQL Server Stretch Database, Cosmos DB, Redis Cache, Table Storage, HDInsight, SQL Data Warehouse, Event Hubs, … Copyright 2018 Expedia Inc. 7
  • 8.
    Container ecosystem may ormay not play a role Copyright 2018 Expedia Inc. 8 Hypothesis
  • 9.
    Are we reallyready? Copyright 2018 Expedia Inc. 9
  • 10.
    Not suitable forlong-running appsExpensive Lock-in Slow on cold start No local storage More complex call patterns Unknown security risks No standards Observability is hard ? Copyright 2018 Expedia Inc. 10
  • 11.
    Habits and inertia Statelessisn’t always stateless Ephemeral isn’t always ephemeral Often pet like ? Copyright 2018 Expedia Inc. 11
  • 12.
    Fear of theecosystem Lock-in worries Undifferentiated heavy lifting vs Value generation? ? Copyright 2018 Expedia Inc. 12
  • 13.
    Maturity Feature gaps Pricing Tomorrow !=Today Tooling? Copyright 2018 Expedia Inc. 13
  • 14.
    Embrace the paradigm Learn Getready Copyright 2018 Expedia Inc. 14

Editor's Notes

  • #12 In-memory config In-memory caches Connection pools Slow startup times Huge runtime Patches
  • #13 In-memory config In-memory caches Connection pools Slow startup times Huge runtime Patches
  • #14 In-memory config In-memory caches Connection pools Slow startup times Huge runtime Patches