... Many IOPS is. Enough? Episode II. Tom Coughlin. Coughlin Associate. &. Jim
Handy. Objective Analysis. © 2012 Coughlin Associates and Objective Analysis ...
How Many IOPS is Enough? Episode II Tom Coughlin Coughlin Associate & Jim Handy Objective Analysis
© 2012 Coughlin Associates and Objective Analysis
Outline • The Survey • Application Distribution and Attributes • More Survey Results: IOPS, Capacity and Latency • Developing tiers of storage for enterprise (and client) applications • Implications/Projections • Authors & Sources © 2012 Coughlin Associates and Objective Analysis
3
Our Survey • Survey is ongoing: TinyURL.com/IOPSsurvey • Very few questions: – – – –
IOPS Capacity Latency Applications
• Results to be published in: – SNIA SSSI white paper of top-level findings – Full in-depth report for sale
© 2012 Coughlin Associates and Objective Analysis
4
Outline • The Survey • Application Distribution and Attributes • More Survey Results: IOPS, Capacity and Latency • Developing tiers of storage for enterprise (and client) applications • Implications/Projections • Authors and Sources 5
Application Types • Cloud Storage/ServiceVirtualization • Databases • On-Line Transaction Processing (OLTP)© • Video Creation and Distribution • Science & Engineering • Exchange Servers 6
Applications from Survey
© 2012 Coughlin Associates and Objective Analysis
7
Cloud Storage/Services-Virtualization Image courtesy of Waring Corp.
• The “IO Blender” – Many streams – Scrambled I/O – Highly random
• Suits SSDs better than HDDs for rapid access • Many VM and VDI systems use flash cache to meet demand speed needs 8
© 2012 Coughlin Associates and Objective Analysis
Databases • • • •
Large data sets Random traffic High I/O load Early SSD adopter – Previously used DRAM SSDs
• Some users load entire DB in flash 9
© 2012 Coughlin Associates and Objective Analysis
OLTP (On-Line Transaction Processing)
• Verified writes – Write/read back – Doubles I/O load
• No room for errors • Speed is imperative – Delays lose customers
10
© 2012 Coughlin Associates and Objective Analysis
Video Creation or Distribution • Large data sets • Multiple video streams – Randomizes access
• High bandwidth required • Expensive talent – Don’t want them sitting around waiting Image courtesy of the US Library of Congress
11
© 2012 Coughlin Associates and Objective Analysis
Recording Media Used in Professional Video Cameras
© 2012 Coughlin Associates and Objective Analysis
12
Science & Engineering Image courtesy of Wikimedia Commons
• Complex problems – – – –
Genome sequencing CAD/CAM Natural resources Nuclear modeling
• Large data sets • Expensive talent – Don’t want them sitting around waiting 13
© 2012 Coughlin Associates and Objective Analysis
Exchange Server • Multiple tasks – e-mail – Scheduling/calenda rs – Data storage
• Scads of users • e-mail chaos – Multiple mailboxes – Asynchronous sends & receives – Spam & virus filters • Strangely we didn’t get any of these users yet Image courtesy of Dell Computer 14
© 2012 Coughlin Associates and Objective Analysis
Outline • The Survey • Application Distribution and Attributes • More Survey Results: IOPS, Capacity and Latency • Developing tiers of storage for enterprise (and client) applications • Implications/Projections • Authors and Sources © 2012 Coughlin Associates and Objective Analysis
15
IOPS Needed by Application
83% between 1 K and 1 M IOPS
© 2012 Coughlin Associates and Objective Analysis
16
Range of Storage Capacity Needs
Possibly a Bi-Modal Distribution 17
IOPS Required vs. Storage Capacity
© 2012 Coughlin Associates and Objective Analysis
18
Maximum IOPS of System Hardware (Before another bottleneck becomes a problem)
91% between 1 K and 1 M IOPS
Good Match to Requirements © 2012 Coughlin Associates and Objective Analysis
19
Minimum Latency of System Hardware (Before another bottleneck becomes a problem)
37% at 10 msec latency
20
IOPS Required vs. Average Storage Latency
© 2012 Coughlin Associates and Objective Analysis
21
Outline • The Survey • Application Distribution and Attributes • More Survey Results: IOPS, Capacity and Latency • Developing tiers of storage for enterprise (and client) applications • Implications/Projections • Authors and Sources © 2012 Coughlin Associates and Objective Analysis
22
Today’s DRAMs: 6,000 Times HDDs’ Speed
From : HDDs and Flash Memory: A Marriage of Convenience, Coughlin Associates and Objective Analysis
© 2012 Coughlin Associates and Objective Analysis
23
107
Memory and Storage Price vs. Bandwidth
1.E+07
106
L1
1.E+06
L2
5 1.E+05
Bandwidth (MB/s)
10
L3
104
1.E+04
DRAM
103
1.E+03
NAND 2 1.E+02
10
101
1.E+01
1.E+00 0
101.E-01-1 10
HDD
Tape
0 1.E+00
3 4 1.E+03 1.E+04 1.E+05 1.E+06 10 10 10 105 106 Price per Gigabyte From Objective Analysis: Are Hybrid Drives Finally Coming of Age? 24 © 2012 Coughlin Associates and Objective Analysis
10
1 1.E+01
10
2 1.E+02
$/GB SSD vs. HDD HDD and Flash are on Parallel Paths $1,000.00 Flash HDD
$100.00 $10.00 $1.00 $0.10 $0.01 2004
2005
2006
2007
2008
2009
2010
Prices: HDD – Price G2, Flash, Objective Analysis 25
© 2012 Coughlin Associates and Objective Analysis
Market Expectations Price/GB Roughly Follows IOPS $10,000.00
$/GB
$1,000.00 $100.00 $10.00 $1.00
Lots of IOPs, Lower Capacity, Higher Cost
Lots of TB’s, Low Cost, Lower IOPs Storage
$0.10 1.E+01
1.E+02
1.E+03
1.E+04
IOPS
26
© 2012 Coughlin Associates and Objective Analysis © 2012 Coughlin Associates and Objective Analysis
1.E+05
1.E+06
Device IOPS by Form Factor
MB/s 27
© 2012 Coughlin Associates and Objective Analysis
Outline • The Survey • Application Distribution and Attributes • More Survey Results: IOPS, Capacity and Latency • Developing tiers of storage for enterprise (and client) applications • Implications/Projections • Authors and Sources © 2012 Coughlin Associates and Objective Analysis
28
Implications/Projections (1) • Increasing SSD adoption – More IOPS is better for most applications
• Fast storage will transform: – Was short-stroked HDDs – Migrating to SSDs
• New enterprise hierarchy – SSD front end – HDD back end 29
© 2012 Coughlin Associates and Objective Analysis
Implications/Projections (2) • Bottleneck is no longer storage – Network, software, servers…
• Users focus more attention on IOPS: – Stronger understanding than today – Higher IOPS drives more data/content • More content means more storage
– Result: widespread storage growth • More SSDs • More HDDs • Even more tape!!! 30
© 2012 Coughlin Associates and Objective Analysis
Participate in the Survey! • SSDs will be priced by speed and capacity – Users will benefit by knowing their speed & capacity needs – Sellers will too!
• Ongoing 5-minute survey at: TinyURL.com/IOPSsurvey • Initial results encouraging • We need more respondents! © 2012 Coughlin Associates and Objective Analysis
UPCOMING REPORT COMPILES IT MANAGER SURVEY RESULTS • Report to be published in late October
Objective-Analysis.com or TomCoughlin.com/techpapers.htm 408-978-8184
[email protected].
© 2012 Coughlin Associates and Objective Analysis
32
Your Presenters Tom Coughlin, President, Coughlin Associates is a highly-respected storage analyst and consultant with over 30 years in the data storage industry in Thomas Coughlin Coughlin Associates
engineering and management at high profile companies. Jim Handy is a widely recognized semiconductor analyst, has over 30 years in the electronics industry.
His background includes marketing and design positions at market-leading suppliers. Jim Handy Objective Analysis © 2012 Coughlin Associates and Objective Analysis
33
Source Documents • How Many IOPS is Enough? – Coughlin Associates & Objective Analysis • HDDs and Flash Memory: A Marriage of Convenience – Coughlin Associates & Objective Analysis • 2012 Digital Storage for Media and Entertainment – Coughlin Associates • Are Hybrid Drives Finally Coming of Age? – Objective Analysis • Two may be Better than One: Why HDD and Flash Belong Together – SNIA SSSI White Paper TomCoughlin.com/techpapers, Objective-Analysis.com
© 2012 Coughlin Associates and Objective Analysis
34