IT professionals find it difficult to determine which SSD or flash array to buy or even whether they can get the speed they need from standard HDDs. There is an extraordinarily wide rage of IOPS (from hundreds to millions), latencies, and capacities, and this can be confusing. This 2016 update to our rigorous 2012 report provides, through a survey of nearly 260 IT managers and other end users, a good understanding of the needs of various applications including IOPS, latency, and capacity.
The 157-page report dissects the results of a months-long on-line survey of IT managers which collected their inputs on the IOPS, capacity, and latency needs for a number of key enterprise applications. Through its 136 figures and eight tables the study sheds new light on the speed requirements for certain key applications like OLTP, database management, and cloud services.
This report details the results of a months-long storage performance survey performed through 2016 compared to earlier responses to the same set of survey questions in 2012.
The survey asked over two hundred sixty respondents to share their high-speed storage capacity and performance needs, IOPS and maximum latency, along with the speed of the rest of their system and for the application type being run on that system.
Data is presented in three graphical formats:
A pie chart to display the survey population
Column charts indicating the percentage of respondents falling into various categories (Pareto Charts)
Scatter charts to help identify the relationship between the different parameters that were measured.
In an effort to determine how perceptions have changed over time, this data is presented for 2012 alone (as it appeared in our original report), through 2016, and with the newer responses broken out by themselves and compared with the older data.
The following results were found:
- The majority of our survey respondents required fast storage for the following application types:
- On-Line Transaction Processing (OLTP)
- Cloud and Storage Services
- Scientific and Engineering Computing
- Those survey responses that changed between 2012 and 2016 generally represented increases in both speed and capacity and reductions in expected latency.
- Most respondents (79%) required performance of over 1,000 but fewer than ten million IOPS over a broad range of capacities
- Latency requirements had a very solid peak with 32% requiring 10 milliseconds.
- In many cases there was a reasonable correlation between two of the surveyed parameters:
- Capacity & IOPS requirements
- Capacity & Maximum Latency
- IOPS Requirements and Maximum Latency
The sample group was a respectable size, giving us a 6.0% margin of error at a 95% confidence level. We continue to collect survey data with a goal of increasing the respondent base and reducing the error margin.
Storage pricing is loosely tied to performance. This relationship is likely to tighten as a result of this survey and as those who purchase fast storage become more sophisticated.
1. Executive Summary
2. Survey Objective
- Why Did We Do This?
- Margin of Error
- Anticipated Outcome
3. Storage Devices and their Characteristics
- SSDs and Flash Memory vs. DRAM
- Caching/Tiering vs. Determinism
4. Raw Data Analysis
- IOPS Requirements
- Capacity Requirements
- Impact of Other System Bottlenecks
- Minimum Useful Latency
- Comparing the Results
- IOPS vs. Capacity
- Latency vs. Capacity
- IOPS vs. Latency
5. Application Sensitivities
- Database Systems
- Metadata Servers
- Data Warehousing and Search
- Internet Server Caching
- On Line Transaction Processing (OLTP)
- Charge Card Processing
- Reservations Systems
- Algorithmic Trading
- Currency Exchange and Arbitrage
- Inter-Bank Transfers
- Other On-Line Transaction Processing Applications
- Cloud or Storage Services
- Real Time Data/Feed Processing
- Contextual Web Advertising
- Scientific or Engineering Applications
- Electronic Design Automation & Modeling
- Weather/Life Sciences
- Aerodynamics Design
- Nuclear Fission Models
- Software Development
6. IOPS Requirements by Application
7. The Monetary Value of Higher IOPS Figures
8. How the Market Should Develop
Appendix A: The Online Survey
Appendix B: Data Inconsistencies
Possible Reasons for Inconsistencies