RADoN: QoS in Storage Networks Andrew Shewmaker 1 Tim Kaldewey 1 - - PowerPoint PPT Presentation

▶
radon qos in storage networks
SMART_READER_LITE
LIVE PREVIEW

RADoN: QoS in Storage Networks Andrew Shewmaker 1 Tim Kaldewey 1 - - PowerPoint PPT Presentation

Oct 03, 2023 169 likes •297 views

RADoN: QoS in Storage Networks Andrew Shewmaker 1 Tim Kaldewey 1 Richard Golding 2 Carlos Maltzahn 1 Theodore M Wong 2 Scott Brandt 1 1 University of California Santa Cruz 2 IBM Almaden Research Center Computer Science Department Storage Systems

slide-1
SLIDE 1

RADoN: QoS in Storage Networks

Tim Kaldewey1 Andrew Shewmaker1 Richard Golding2

Carlos Maltzahn1

Theodore M Wong2

Scott Brandt1

1University of California Santa Cruz

Computer Science Department {kalt,shewa,carlosm,scott} @cs.ucsc.edu

2IBM Almaden Research Center

Storage Systems Department {rgolding,theowong} @us.ibm.com

This work was supported by NSF Award No. CCF-0621534

slide-2
SLIDE 2

2

Storage QoS - XXL

.

End-to-End storage QoS

slide-3
SLIDE 3

3

Storage QoS in practice

.

Common approach: phys. partitioning to achieve isolation End-to-End storage QoS

slide-4
SLIDE 4

4

Storage QoS in practice

.

Common approach: phys. partitioning to achieve isolation

  • ften  overprovisioning =(

End-to-End storage QoS

slide-5
SLIDE 5

5

Storage QoS in research

.

End-to-End storage QoS Research on: Caching Network QoS Disk scheduling

slide-6
SLIDE 6

6

Storage QoS in research

.

End-to-End storage QoS Research on: Caching Network QoS Disk scheduling But no integration =(

slide-7
SLIDE 7

7

End-to-End storage QoS ?

.

End-to-End storage QoS How does guaranteed storage performance translate to network and cache requirements? How to coordinate network, cache and storage subsystems?

slide-8
SLIDE 8

8

End-to-End storage QoS ?

  • Large storage system  Many parameters to tweak
  • Which are important?
slide-9
SLIDE 9

9

RADoN – model

  • Large storage system  Many parameters to tweak
  • Which are important?  Find out via simulation:
slide-10
SLIDE 10

10

RADoN – simulation results

.

Time series for throughput of 4 clients, each reserving 25% of storage performance, but producing enough results to saturate the disk itself

100% = Disk 25% = Disk

slide-11
SLIDE 11

11

Work in Progress

 Simulating multiple approaches to coordinate subsystems  Implementation on top of existing QoS disk scheduler [Fahrrad]  Complete E2E storage QoS framework [RADI/O]

slide-12
SLIDE 12

12

Work in Progress

 Simulating multiple approaches to coordinate subsystems  Implementation on top of existing QoS disk scheduler [Fahrrad]  Complete E2E storage QoS framework [RADI/O]

Long term goal: better storage QoS to avoid