VAT: Asymptotic Cost Analysis for Multi-Level Key-Value Stores - - PowerPoint PPT Presentation

▶
vat asymptotic cost analysis for multi level key value
SMART_READER_LITE
LIVE PREVIEW

VAT: Asymptotic Cost Analysis for Multi-Level Key-Value Stores - - PowerPoint PPT Presentation

Mar 21, 2024 173 likes •308 views

VAT: Asymptotic Cost Analysis for Multi-Level Key-Value Stores Graduate Students Conference Nikos Batsaras 25 October 2019 Computer Science Department, University of Crete Key-Value (KV) Store Application data Client KV store :

slide-1
SLIDE 1

VAT: Asymptotic Cost Analysis for Multi-Level Key-Value Stores

Graduate Students Conference

Nikos Batsaras 25 October 2019

Computer Science Department, University of Crete

slide-2
SLIDE 2

Key-Value (KV) Store

  • KV store: Dictionary of KV pairs.
  • API: put, get, scan, delete.
  • Use cases: web indexing, social

networks, data analytics.

  • Storage engine for many large scale

storage systems.

  • Google (LevelDB)
  • Facebook, Yahoo! (RocksDB)
  • Microsoft (FASTER)
  • Apple (FoundationDB)
  • . . .

Application data

(k,v) Client Server KV store kv kv kv kv · · ·

Main memory

kv kv kv kv · · ·

Persistent device

nikbats@ics.forth.gr 1 of 8

slide-3
SLIDE 3

Persistent Multi-Level KV Stores

Definitions:

  • SST: Sorted String Table.
  • Si: Capacity of level Li.
  • f: Growth factor.

Main memory Persistent device

Compaction S0

KV · · · KV · · · SST

L0 S1 = f · S0

KV · · · KV SST

L1

KV · · · KV · · · SST

. . .

Sℓ = f · Sℓ−1

KV · · · KV SST

Lℓ

KV · · · KV SST KV · · · KV · · · SST

nikbats@ics.forth.gr 2 of 8

slide-4
SLIDE 4

Persistent Multi-Level KV Stores

Definitions:

  • SST: Sorted String Table.
  • Si: Capacity of level Li.
  • f: Growth factor.

Design choices:

  • What growth factor to pick ?
  • What should the SST size be ?
  • What compaction algorithm to use ?

Main memory Persistent device

Compaction S0

KV · · · KV · · · SST

L0 S1 = f · S0

KV · · · KV SST

L1

KV · · · KV · · · SST

. . .

Sℓ = f · Sℓ−1

KV · · · KV SST

Lℓ

KV · · · KV SST KV · · · KV · · · SST

nikbats@ics.forth.gr 2 of 8

slide-5
SLIDE 5

Problem Statement & Challenge

  • Goal: Optimize KV-store configuration.
  • Problem: Complexity of design space.
  • 1. Many different designs that optimize for

different metrics.

  • 2. New device technology further complicates

the design space.

  • Challenge: An analysis that guides towards
  • ptimal configuration.

Designer Analysis

I/O Amplification Growth factor Compaction SST size Optimal KV store

nikbats@ics.forth.gr 3 of 8

slide-6
SLIDE 6

Landscape

I/O Amplification Storage Evolution HDD SSD/NVMe NVM

nikbats@ics.forth.gr 4 of 8

slide-7
SLIDE 7

Landscape

I/O Amplification Storage Evolution HDD SSD/NVMe NVM

Amplification with technology trend

RocksDB Dostoevsky Monkey PebblesDB SifrDB Atlas WiscKey HashKV Kreon ?

nikbats@ics.forth.gr 4 of 8

slide-8
SLIDE 8

Landscape

I/O Amplification Storage Evolution HDD SSD/NVMe NVM VAT

Amplification with technology trend

RocksDB Dostoevsky Monkey PebblesDB SifrDB Atlas WiscKey HashKV Kreon ?

nikbats@ics.forth.gr 4 of 8

slide-9
SLIDE 9

Experimental Results

2 4 8 64

Growth factor (f)

10 20 30 40 50 60

Cost RocksDB approximation

RocksDB SST-VAT VAT 2 4 8 64

Growth factor (f)

0.0 0.5 1.0 1.5 2.0

Cost Kreon approximation

VAT Kreon 2 4 8 64

Growth factor (f)

0.0 0.5 1.0 1.5 2.0

Cost BlobDB approximation

VAT BlobDB 2 4 8 64

Growth factor (f)

2 4 6 8 10

Cost PebblesDB approximation

VAT PebblesDB nikbats@ics.forth.gr 5 of 8

slide-10
SLIDE 10

Single Tier for Future Fast Storage Devices

  • Paper: Basic performance measurements of the intel optane DC

persistent memory module. CoRR, abs/1903.05714, 2019.

  • Maximum throughput with 256-byte request size in NVM.
  • VAT says that NVM-based KV stores will use 1 device resident level.

Main memory NVM device

Compaction S0

KV · · · KV · · · SST

L0 Dataset

KV · · · KV SST

L1

KV · · · KV · · · SST

nikbats@ics.forth.gr 6 of 8

slide-11
SLIDE 11

Conclusions

In this presentation, we talked about the VAT analysis which:

  • 1. Expresses I/O cost for different KV-store designs.
  • 2. Captures the technology impact on KV-store design.
  • 3. Quantifies tradeoffs between different design decisions.
  • 4. Guides towards optimal configuration.
  • 5. Makes future projections based on technology trends.

nikbats@ics.forth.gr 7 of 8

slide-12
SLIDE 12

Questions ?

Thank you.

Institute of Computer Science, FORTH – Heraklion, Greece Computer Science Department, University of Crete – Heraklion, Greece

  • Email: nikbats@ics.forth.gr
  • Web: http://www.ics.forth.gr/carv

nikbats@ics.forth.gr 8 of 8