CSD 2000 SERIES Dramatically Reduce Storage Costs in PostgreSQL ScaleFux Computational Storage Drives (CSD) dramatically improve database cost efficiency, granting much lower $/bit versus NVMe SSDs. PostgreSQL does not natively support data compression. For every one GB of ordinary SSD storage, users can store at most one GB of PostgreSQL data. However, the CSD 2000 compresses data on writes, allowing users to store up to 4x as much data, delivering a 75% reduction in storage cost (Figure 1). This improvement is even more dramatic when admins want to boost PostgreSQL performance by setting a lower Fill Factor (see below). CSD 2000 provides Penalty-Free Compression • Perform compression/decompression in the drive, without burdening the CPU and without adding Latency • Scale throughput with each drive and avoid CPU bottlenecks • Improve storage efficiency: lower cost to store each byte of data © 2020 ScaleFlux, Inc. All rights reserved. ScaleFlux is a registered trademark of ScaleFlux, Inc. Other names and brands may be claimed as the property of others. ScaleFlux, Inc. 97 East Brokaw Rd, Ste 260 San Jose, CA 95112 www.scaleflux.com SUPERCHARGE YOUR POSTGRESQL INFRASTRUCTURE! REQUEST A POC: [email protected] ScaleFlux ® CSD 2000 Series - - PCIe AIC & U.2 Drive - up to 20TB user capacity (8TB physical) - PCIe Gen3 x4 - Form Factor Capacity Interface Storage Cost Lower is Better 0 1.0 1.5 Storage Capacity Comsumption (TB) (Lower is Better) 0.5 Fill Factor 100 Fill Factor 75 NVMe SSD CSD 2000 6KB Data 2KB 0ʼs Compressed Data Fill Factor = 75; 8KB/Page Data Path Compression Fill Factor (FF) is the percent of each page that is filled with user data and can be configured to trade off storage cost and performance. Higher FF = slower but cheaper; Lower FF = faster but more expensive. Storage Cost -- The default FF setting of 100% is optimized for storage cost efficiency. Reducing FF to 75%, PostgreSQL initially fills each 8KB page with 6KB of user data and 2KB of space reserved for future updates (Figure 2). The reserved space is filled with highly-compressible data (all zeros). Ordinarily, inserting reserved space directly increases storage footprint and cost. Database Performance -- Lower FF yields better higher PostgreSQL performance (Figure 3). Our simulated TPC-C workload saw a 35% increase in QPS when Fill Factor was set to 75%. CSD 2000 enables users to gain the QPS benefits of lower FF without incurring its traditional drawbacks, actually achieveing a 75-80% reduction in storage cost vs an ordinary SSD. Fill Factor in PostgreSQL - Trading Storage Cost for Database Performance Lower Flash Cost Sysbench TPC-C 75 % large TCO SAVINGS Figure 1 Figure 2 0 100 Queries per Second (QPS) (Higher is Better) 125 75 50 Fill Factor 100 Fill Factor 75 25 Figure 3 2.0 Sysbench TPC-C QPS Higher is Better Configuration information for 1.4TB Sysbench TPC-C test: Fill Factor 100: raw data set size 1.4TB (Compressed to 341GB on CSD 2000) Fill Factor 75: raw data set size 1.4TB (Compressed to 364GB on CSD 2000)
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
![Page 1: Dramatically Reduce Storage Costs in PostgreSQL TCO SAVINGS200113]ScaleFlux... · 2020-02-13 · PostgreSQL does not natively support data compression. For every one GB of ordinary](https://reader033.cupdf.com/reader033/viewer/2022050401/5f7eeacaadf8a26132468269/html5/thumbnails/1.jpg)
CSD 2000 SERIES
Dramatically Reduce Storage Costs in PostgreSQL
ScaleFux Computational Storage Drives (CSD) dramatically improve database cost efficiency, granting much lower $/bit versus NVMe SSDs.
PostgreSQL does not natively support data compression. For every one GB of ordinary SSD storage, users can store at most one GB of PostgreSQL data. However, the CSD 2000 compresses data on writes, allowing users to store up to 4x as much data, delivering a 75% reduction in storage cost (Figure 1). This improvement is even more dramatic when admins want to boost PostgreSQL performance by setting a lower Fill Factor (see below). CSD 2000 provides Penalty-Free Compression • Perform compression/decompression in the drive, without burdening the CPU and without adding Latency • Scale throughput with each drive and avoid CPU bottlenecks • Improve storage efficiency: lower cost to store each byte of data
© 2020 ScaleFlux, Inc. All rights reserved. ScaleFlux is a registered trademark of ScaleFlux, Inc.Other names and brands may be claimed as the property of others.
ScaleFlux, Inc.97 East Brokaw Rd, Ste 260
San Jose, CA 95112www.scaleflux.com
SUPERCHARGE YOUR POSTGRESQL INFRASTRUCTURE! REQUEST A POC: [email protected]
ScaleFlux® CSD 2000 Series
- - PCIe AIC & U.2 Drive - up to 20TB user capacity (8TB physical) - PCIe Gen3 x4
-
Form Factor Capacity Interface
Storage CostLower is Better
0
1.0
1.5
Stor
age
Cap
acity
Com
sum
ptio
n (T
B)(L
ower
is B
ette
r)
0.5
Fill Factor 100 Fill Factor 75
NVMe SSD CSD 2000
6KB Data 2KB 0ʼs
Compressed Data
Fill Factor = 75; 8KB/Page
Data PathCompression
Fill Factor (FF) is the percent of each page that is filled with user data and can be configured to trade off storage cost and performance. Higher FF = slower but cheaper; Lower FF = faster but more expensive. Storage Cost -- The default FF setting of 100% is optimized for storage cost efficiency. Reducing FF to 75%, PostgreSQL initially fills each 8KB page with 6KB of user data and 2KB of space reserved for future updates (Figure 2). The reserved space is filled with highly-compressible data (all zeros). Ordinarily, inserting reserved space directly increases storage footprint and cost. Database Performance -- Lower FF yields better higher PostgreSQL performance (Figure 3). Our simulated TPC-C workload saw a 35% increase in QPS when Fill Factor was set to 75%. CSD 2000 enables users to gain the QPS benefits of lower FF without incurring its traditional drawbacks, actually achieveing a 75-80% reduction in storage cost vs an ordinary SSD.
Fill Factor in PostgreSQL -Trading Storage Cost for Database Performance
Lower Flash CostSysbench TPC-C75%
large TCO SAVINGS
Figure 1
Figure 2
0
100
Que
ries
per
Seco
nd (
QPS
)(H
ighe
r is
Bette
r)
125
75
50
Fill Factor 100 Fill Factor 75
25
Figure 3
2.0
Sysbench TPC-C QPSHigher is Better
Configuration information for 1.4TB Sysbench TPC-C test:Fill Factor 100: raw data set size 1.4TB (Compressed to 341GB on CSD 2000)Fill Factor 75: raw data set size 1.4TB (Compressed to 364GB on CSD 2000)
Related Documents
