lightning talk: turbo charge your mongodb app
DESCRIPTION
High performance systems at scale can challenge storage infrastructures. Today, many MongoDB customers are addressing this by moving to all-flash shared storage platforms, which can dramatically reduce latency. Hear from the experts at enterprise storage provider Pure Storage on how 3 leading online SaaS providers have improved performance, become more efficient, and simplified their MongoDB infrastructure with all-flash storage.TRANSCRIPT
© 2014 Pure Storage, Inc. | 1 Pure Storage CONFIDENTIAL – Do Not Distribute
Introducing Pure StorageTurbo Charge your MongoDB App
Markus GrauNovember, 2014
© 2014 Pure Storage, Inc. | 2 Pure Storage CONFIDENTIAL – Do Not Distribute
Consistent performance100% MLC Flash
Less cost than diskInline deduplication & compression
Mission-critical reliability99.999%, non-disruptive operations
Pure Storage is Flash for the Enterprise
Scalable & compatible10 100s of TBs, Purity software
© 2014 Pure Storage, Inc. | 3 Pure Storage CONFIDENTIAL – Do Not Distribute
IBM24.6%
Pure Storage17.1%
Violin13.2%
EMC11.1%
NetApp10.6%
Cisco3.2%
SolidFire3.1%
HP1.3%
Others15.8%
– Pure Storage #2 Solid State Array Market ShareFastest Growing All-Flash Leader
Data Source: Gartner Research, June 10, 2014 Chart: Pure Storage
182%SSA Market
Growth
Gartner Market Share Analysis: Solid State Arrays Worldwide 2013
Others24%
IBM18%
Pure Storage7%
Violin30%
NetApp13%
Cisco8%
2012$236M
2013$667M
#2Market Share
>600%YoY Growth
20132013
© 2014 Pure Storage, Inc. | 4 Pure Storage CONFIDENTIAL – Do Not Distribute
Pure Storage FlashArray 400 Series
FA-405 FA-450FA-420
Up to 40TBs2 controllers in
2U
Up to 250TBs16 Gb/s FC
Up to 125 TBs
Pricing includes deduplication, snapshots, replication, etc.
Usable Capacity includes overheads for RAID-3D, Flash Management, GB-to-GiB conversion, and assumes 6-to-1 data reduction.
© 2014 Pure Storage, Inc. | 5 Pure Storage CONFIDENTIAL – Do Not Distribute
Typical Scaling Concerns for MongoDB Admins
•Performance• Disk performance has not changed (GB/IOP is getting worse)• SSD is expensive at scale
•Complexity• Scaling increases complexity (sharding, caching, tiering, etc)• Requires significant management / training• Mistakes = Downtime or Performance Impact
•Inefficiency• Underutilized capacity • Data center power, cooling, rack space is expensive
© 2014 Pure Storage, Inc. | 6 Pure Storage CONFIDENTIAL – Do Not Distribute
Performance: Results running MongoDB
© 2014 Pure Storage, Inc. | 7 Pure Storage CONFIDENTIAL – Do Not Distribute
Case Study: Cengage Learning •Profile
• Educational Services Provider offering wide variety of online services worldwide
• Run analytics, data warehouse, and royalty apps based on MongoDB
•Performance• “(after replacing disk with Pure) “In almost all cases, a .5 ms response time
was a 500-700% improvement in data store response time”
•Simplicity• "The Pure array has done everything as advertised”
•Efficiency• "Pure has brought to light major configuration issues we had in our own
environment -- some with multi-pathing, some of our UCS configuration”
- Jason Spruance, Operations Manager
© 2014 Pure Storage, Inc. | 8 Pure Storage CONFIDENTIAL – Do Not Distribute
Reducing Complexity: Configuring Pure for MongoDB
Provision a LUN of appropriate size for your DB• No need to create RAID groups• No need to set cache policies
Pure is block storage, so access from compute nodes:• Via Ethernet for iSCSI• Via dedicated Fiber Channel Network
Create zero space DB clones in seconds
© 2014 Pure Storage, Inc. | 9 Pure Storage CONFIDENTIAL – Do Not Distribute
ND Non-Disruptive Everything*For a true enterprise consolidation array hosting 100s of
applications, downtime is simply NOT an option.
Non-DisruptiveCapacity Expansion
Non-DisruptivePerformance
Expansion
Non-DisruptiveHardware
Replacement
Non-DisruptiveSoftware Updates
…All WithZero Performance
Loss
© 2014 Pure Storage, Inc. | 10 Pure Storage CONFIDENTIAL – Do Not Distribute
Case Study: Online Photo Company•Profile
• Rapid growth taxing traditional disk storage
•Performance• Move to Pure resulted in “Half the average response time”• 4x faster for replication to catch up after maintenance
•Simplicity• Array Installed in an hour• No configuration of devices • Mongo Team Manages Their Own Storage
•Efficiency:• Averaging 12 to 1 Data Reduction• 16TB Production Mongo Instance stored in 1.35TB of Pure Storage
© 2014 Pure Storage, Inc. | 11 Pure Storage CONFIDENTIAL – Do Not Distribute
Efficiency: Pure Storage Data Reduction
• Comprehensive, high-performance data reduction
• Pattern Removal • Adaptive, Inline Deduplication • Adaptive, Inline Compression • Deep Reduction • Zero-cost snapshots, clones & XCopy
• 5 – 10x reduction across broad number of data sets
• Reduces usable €/GB
* Based on CloudAssist data of Pure FlashArray deployments as of April 2014
© 2014 Pure Storage, Inc. | 12 Pure Storage CONFIDENTIAL – Do Not Distribute
Typical Customer Data Reduction Results
© 2014 Pure Storage, Inc. | 13 Pure Storage CONFIDENTIAL – Do Not Distribute
The Flash Price Barrier Has Been Broken!
$/GBHA
USABLEW/ SW
2012 2013 2014 2015 2016 2017…
$6$5 $4 10/15K
HDD
$2 $1.5 $1 SATAHDD
>$20
$15
$10
$6
$3
FlashAppliances
$8
$5
$4
$2$1
5-10xData Reduction!
Consumer SSD + Data Reduction
= Cost Parity to Disk
© 2014 Pure Storage, Inc. | 14 Pure Storage CONFIDENTIAL – Do Not Distribute
Typical Pain Points: Addressed
•Performance• All-flash combines benefits of SAN with SSD performance
•Simplicity• Pure’s all-flash implementation is radically simpler than disk or all-flash alternatives
•Efficiency• Data Reduction maximizes utilization
© 2014 Pure Storage, Inc. | 15 Pure Storage CONFIDENTIAL – Do Not Distribute
Case Study: Healthcare SaaS Provider•Profile
• Runs their business on applications leveraging Mongo DB
• < 1 ms latency critical for end user experience.
•Performance• Performance so good that customer will use Pure as a sales tool to attract
customers and differentiate from competition
•Simplicity• Pure allows their Network Engineer to be their storage expert & continue growth
without additional FTE headcount
•Efficiency• 5:1 data reduction for MongoDB enables scale without sprawl• Will consolidate a HP & DELL array into Pure for production
© 2014 Pure Storage, Inc. | 16 Pure Storage CONFIDENTIAL – Do Not Distribute
MongoDB Scaling Recommendations
Compute Nodes
DB Size
1
2
4
8
1TB 2TB 4TB 8TB 16TB
Use Commodity Servers with PCIe Flash:
• Low node count • Low application
count• Small databases
Migrate to All-flash SAN when:
• Multiple nodes increase complexity
• Database size taxes disk I/O
Upgrade:• Servers based on
performance requirements and CPU cycles
• Storage based on capacity requirements
© 2014 Pure Storage, Inc. | 17 Pure Storage CONFIDENTIAL – Do Not Distribute
Reference Architecture
•Scalable Capacity
•Resiliency
•Consistent Performance
•Data Reduction
•Ease of Management
•Snapshots and Database Cloning
http://info.purestorage.com/WP-PureStorageandMongoDB_Request.html
© 2014 Pure Storage, Inc. | 18 Pure Storage CONFIDENTIAL – Do Not Distribute
Thank You! CONFIDENTIAL – DO NOT DISTRIBUTE