Can Microservices Drive a Renaissance in Workload-Aware Storage - - PowerPoint PPT Presentation

can microservices drive a renaissance in workload aware
SMART_READER_LITE
LIVE PREVIEW

Can Microservices Drive a Renaissance in Workload-Aware Storage - - PowerPoint PPT Presentation

Can Microservices Drive a Renaissance in Workload-Aware Storage Management? 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage 20) Pranav Bhandari 1 , Lukas Rupprecht 2 , Dimitrios Skourtis 2 , Ali Anwar 2 , Deepavali


slide-1
SLIDE 1

Can Microservices Drive a Renaissance in Workload-Aware Storage Management?

1

12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage ’20) Pranav Bhandari1, Lukas Rupprecht2, Dimitrios Skourtis2, Ali Anwar2, Deepavali Bhagwat2, Vasily Tarasov2, Avani Wildani1

1Emory University, 2IBM

slide-2
SLIDE 2

Microservices

2 Gan, Yu, et al. "An open-source benchmark suite for microservices and their hardware-software implications for cloud & edge systems." Proceedings of the Twenty-Fourth International Conference on Architectural Support for Programming Languages and Operating Systems (ASPLOS). 2019.

➔ Isolation ➔ Flexibility ➔ Productivity ➔ Scalability ➔ Storage?

slide-3
SLIDE 3

Storage

3

A B C

Distributed Storage (OpenEBS, Ceph, GPFS, GlusterFS)

STORAGE APP

slide-4
SLIDE 4

Microservices

Research

4

Monolithic

A B C

Workload Characterization

Distributed Storage

Auto-tuning

slide-5
SLIDE 5

Microservices

Motivation

5

Monolithic

A B C

Monolithic Microservices A heterogeneous mix of workloads from different application components A set of workloads from application components (services) that each perform a simple task Shared storage Each microservice can have its own volume which can be provisioned dynamically DFS

slide-6
SLIDE 6

Workload Stability

6

➔ Access pattern based workload metrics

◆ read/write ratio ◆ locality ◆ I/O size distribution

➔ Are these metrics more stable in the workloads of microservices compared to the monolithic workload of functionally similar application?

slide-7
SLIDE 7

Storage Auto-tuning

7

➔ Storage parameters

◆ cache (size, write policy, replacement policy, prefetching) ◆ replication ◆ block size ➔ Case study: Cache Size

slide-8
SLIDE 8

Setting

8

➔ Isolated I/O cache in host memory

◆ Local in-memory data access ◆ No network request to the storage service ➔ Cache allocated per persistent volume mounted

  • n the host

◆ Size cache based on the workload of the persistent volume

slide-9
SLIDE 9

Shared vs Isolated Cache

9

Isolated cache performs better than shared cache!

slide-10
SLIDE 10

Cache Size Allocation

10

Workload analysis is needed for cache allocation!

slide-11
SLIDE 11

Initial Design

11

slide-12
SLIDE 12

Thank You!

12

Can Microservices Drive a Renaissance in Workload-Aware Storage Management?

Pranav Bhandari1, Lukas Rupprecht2, Dimitrios Skourtis2, Ali Anwar2, Deepavali Bhagwat2, Vasily Tarasov2, Avani Wildani1

1Emory University, 2IBM

pranav.bhandari@emory.edu