The SPEC Cloud Group's Research Vision on FaaS and Serverless - - PowerPoint PPT Presentation

the spec cloud group s research vision on faas and
SMART_READER_LITE
LIVE PREVIEW

The SPEC Cloud Group's Research Vision on FaaS and Serverless - - PowerPoint PPT Presentation

The SPEC Cloud Group's Research Vision on FaaS and Serverless Architectures Erwin van Eyk (TU Delft) Alexandru Iosup (VU / TU Delft) Simon Seif (SAP SE) Markus Thmmes (IBM) 1 SPEC RG Cloud - Serverless Gaining deeper understanding in


slide-1
SLIDE 1

1

The SPEC Cloud Group's Research Vision on FaaS and Serverless Architectures

Erwin van Eyk (TU Delft) Alexandru Iosup (VU / TU Delft) Simon Seif (SAP SE) Markus Thömmes (IBM)

slide-2
SLIDE 2

2

SPEC RG Cloud - Serverless

Gaining deeper understanding in serverless and FaaS architectures, with a focus on performance (evaluation).

slide-3
SLIDE 3

Cloud (Native) Application

3

Business Logic vs. Operational Logic

Types of logic FaaS Serverless Challenges

slide-4
SLIDE 4

4

Business Logic vs. Operational Logic

Operational Logic Business Logic Logic directly related to use case:

  • Fetching a user’s balance
  • Generating a daily report
  • Calculating portfolio risk

Logic related to QoS:

  • Keeping OS up to date
  • Serving clients
  • Managing DB connections

Types of logic FaaS Serverless Challenges

slide-5
SLIDE 5

5

Joint Problem

Operational Logic Business Logic

Cloud users: want to avoid complex operational logic Cloud providers: aim for higher resource utilization

Types of logic FaaS Serverless Challenges

slide-6
SLIDE 6

6

Monoliths

  • Difficult to Scale
  • Infrequent, complex

deployments

  • Tightly coupled stack

Monolithic Application Operational Logic Infrastructure

Monoliths Microservices FaaS

Types of logic FaaS Serverless Challenges

slide-7
SLIDE 7

7

Microservices (µs)

  • scalable
  • DevOps practices
  • Complexity shifts from

application logic to

  • perational logic.

µs Operational Logic Infrastructure µs Operational Logic µs Operational Logic µs Operational Logic

Monoliths Microservices FaaS

Types of logic FaaS Serverless Challenges

slide-8
SLIDE 8

Types of logic FaaS Serverless Challenges

8

Function-as-a-Service

  • Clear separation of business

logic vs. operational logic

  • Minimal unit of deployment
  • Minimal coupling between each

layer

Operational Logic Infrastructure Function Function Function Function Function Function Function Function

Monoliths Microservices FaaS

slide-9
SLIDE 9

9

Serverless vs. FaaS

Serverless

  • (Almost) no operational logic
  • Event-Driven
  • Granular billing

FaaS

  • A form of serverless computing
  • User provides a function —

deployed and managed by cloud provider Types of logic FaaS Serverless Challenges

slide-10
SLIDE 10

10

What is next in Serverless?

Types of logic FaaS Serverless Challenges

slide-11
SLIDE 11

11

Further Separation of Business and Operational Logic

  • Function Composition
  • Serverless workflows

Function A Function C Function B Function D Types of logic FaaS Serverless Challenges

slide-12
SLIDE 12

12

Focus on Cost/Performance

  • Non-Functional Requirements (NFRs)

priority: high security: max scheduler priority: low security: public

$$$ $

Types of logic FaaS Serverless Challenges

slide-13
SLIDE 13

13

Hybrid Clouds

  • Benchmark of FaaS platforms
  • Reliability
  • Latency: cold/hot starts
  • Throughput

Types of logic FaaS Serverless Challenges

slide-14
SLIDE 14

14

Roadmap

  • Extended vision
  • Reference Architecture
  • Benchmark
slide-15
SLIDE 15

15

HotCloudPerf’18

  • “Performance in the cloud datacenter”
  • April 9, 2018 in Berlin, Germany
  • https://hotcloudperf.spec.org/
  • Held in conjunction with ICPE 2018
slide-16
SLIDE 16

16

Interested?

https://research.spec.org/working-groups/rg-cloud.html @erwinvaneyk E.vanEyk@atlarge-research.com


🌑

slide-17
SLIDE 17

17

Additional Slides

slide-18
SLIDE 18

18

Ongoing work: reference architecture for FaaS platforms

slide-19
SLIDE 19

19

Why Research Serverless and FaaS?

  • Growing industry-driven adoption.
  • Current approaches are still very

immature and wasteful.

  • Far more logic delegation to the

infrastructure (us!).

  • New technologies, same issues
  • rchestration, versioning, scheduling, testing,

monitoring, benchmarking…