TidalScale Overview

Scale | Simplify | Optimize | Evolve
RESTORING DEVELOPER PRODUCTIVITY THROUGH SIMPLICTY
3/21/2016 COPYRIGHT 2014 TIDALSCALE, INC. 1

Industry Context
Data sizes for data under management are monotonically increasing
◦ Who wants less data?
Our appetite for analysis is monotonically increasing
◦ Do you think, or do you know?
◦ Trend toward evidence-based management
Our appetite for speed is monotonically increasing
◦ Who wants questions answered more slowly?
◦ Hence the industry interest in in-memory data management systems
Our overall ability to manage complexity is not increasing

Virtualization today:
Is about Single Node Resource Sharing
Solution scaling constrained by physical machine cost and size
Result: Finite resource pool

Scale-Out today:
An Architecture of Developer Complexity
3/21/2016
COPYRIGHT 2014 TIDALSCALE, INC. 4
Solution scaling is constrained by the programmers ability to comprehend and
navigate the architectural complexity
Result: Increased development cost

TidalScale enables:
An Architecture of Simplicity – One Operating System
3/21/2016
Solution scaling enabled by aggregating hardware resources via one OS instance
Result: Lower development complexity, lower development cost

Why?
Creates a unique scalable solution experience:
User experience bare metal User experience TidalScale
Real Screen Shots
No API’s to learn – define the machine size - boot it up

Fault Tolerance
System Standby and Linux Tool Transparency
3/21/2016
Active Server Standby Server
Transparent access to Linux Availability Tool Chain
TidalScale handles the underlying complexity

Result
AnyLogic™ 1 Million Agent Simulation*
Normal run time - 3 Days
TidalScale run time - 30 minutes
We changed the customers approach to using AnyLogic™
and rapid prototyping of simulations
*No code changes were required

What Kinds of Problems Benefit?
Data Mining / Finance
◦ High Data Volumes, Large Analytics, Risk Analysis, Fraud Detection, Graph Analytics using
Alternative Data Sources, Risk modeling, High Frequency trading, Complex Event Modeling
Bioinformatics
• Next Generation DNA sequencing, Meta-genomic analysis, Finite Element Brain Modeling,
Time-Series MRI Neuro-Imaging
IT / Operational Systems
◦ In-House Applications, Web Controllers & Servers, Gateways, Image serving, Ad serving, OLTP,
ERP, Business Intelligence

Data Mining: High Data Volumes
Scenario
◦ Analytic app running on popular SQL database
◦ Well established – trained users, tied to diverse apps & tools
◦ Running well on a single system, except…
Problem
◦ …Data volumes steadily growing…
◦ …Frequently upgrading hardware…
◦ …About to hit a wall
The TidalScale Solution
◦ Scale up existing application & move more data into memory cache for real time performance
◦ Avoid re-architecting DB & rewriting applications for scale out/cluster
◦ Avoid expensive maintenance cost of a cluster as shape of data changes or use cases & query types evolve

Data Mining: Large Scale Analytics
Scenario
◦ Analytic app running on popular SQL database
◦ Well established – trained users, tied to diverse apps & tools
◦ Proven, reliable application needing no changes, other than the need to increase capacity
Problem
◦ Number of users and connections is growing
◦ Each user is generating large, demanding queries
◦ Frequently upgrading hardware
◦ About to hit a wall – add a second, third copy and offload users?
◦ But then you have a copy/synch problem, especially if data ever gets updated
The TidalScale Solution
◦ Assemble a bigger computer to move more data into memory cache & take advantage of more CPUs
◦ Avoid re-architecting DB for scale out/cluster
◦ Avoid administering multiple servers & users, and copying/synching data

Data Mining: Mortgage Risk Analysis
Scenario
◦ New risk assessment tools for loans, mortgages
◦ Use of “alternative data” in addition to credit bureaus
Problem
◦ Mortgage application workload gets “peaky” as consumers react & play the interest rate game
◦ Backlogs develop during peaks, loan processing still remarkably hands-on process. When backlogs build, customers
leave
TidalScale Solution
◦ Enables a big, flexible computer that scales to handle more complex processing
◦ Supports processor intensive analytics of unstructured data, example NLP (natural language processing)
◦ Easy system expansion as data volumes grow from the addition of unstructured data

Data Mining: Fraud Detection
Scenario
◦ New types of tools to detect fraud
◦ Ex: Graph DBs to see coordinated activity from disparate data
Problem
◦ Graph databases inherently require a closely coherent (single system) view
◦ As system grows, you must either
◦ Split the graph – at the risk of losing potential connections – leaves places for bad guys to hide
◦ Reduce granularity of data – abstracting your view – loss of detail
TidalScale Solution
◦ Enable full detailed view of transactional and account data across all customers over time
◦ Run the graph in-memory for real-time performance
◦ Run system on entirely standard hardware, OS & operational infrastructure

What enables the solution?
The Memory Hierarchy in Human Terms
(.3ns = 1s)
Event Latency Scaled
1 CPU Cycle 0.3 ns 1 s
Level 1 Cache Access 0.9 ns 3 s
Main Memory Access (DRAM, from CPU) 50.0 ns 3 min
Memory over Ethernet 3.2 μs 3.2 hours
CPU Context State Transfer 6.0 μs 6.0 hours
Flash SSD (PCI-e) 4.7 ms 5 months
Rotational disk I/O 1-10 ms 1-12 months
Internet: San Francisco to New York 40 ms 4 years
Internet: San Francisco to United Kingdom 81 ms 8 years
Internet: San Francisco to Australia 183 ms 19 years
TCP packet retransmit 1-3 s 105-317 years

TidalScale
Completely Transparent - No Changes Required
…and many other applications

Hardware Example
System features
• Admin node 1
• Worker nodes 25
• Total Memory 3.2TB
• Total Cores: 150
• Network 1/10GbE
• Storage FreeNAS, xTB
Components
• 1x Admin node (A003)
• Colfax CX1260i-X6 Haswell, E5-2603V3 6C, 16GB
• 25x Worker nodes
• Colfax CX1260i-X6 Haswell, E5-2603V3 6C, 128GB
• 1x 1G switch
• 2x 10G switch (S009, S010) Mellanox
• 1x NAS
163/21/2016 COPYRIGHT 2014 TIDALSCALE, INC.

TidalScale Benefits Summary
In-Memory Performance
◦ The world wants data to be in-memory, but hasn’t been able to get it.
◦ Historically the industry has scaled applications to fit on available computer hardware. Now, for the first time, the
industry can scale the hardware to fit the application.
Linear System Scalability
◦ TidalScale makes it possible to organically grow hardware using low cost commodity servers at linear cost, as
customer needs evolve.
◦ Applications can now achieve superior in-memory performance using inexpensive unmodified hardware.
Reduced Software Development Costs
◦ TidalScale uses off-the-shelf, unmodified Linux.
◦ TidalScale requires no changes to applications or database software.
◦ Optimization happens automatically using our software, and learns over time.

Price/Performance at Scale
Scale up costs increase
rapidly after 3Tb…

Performance Scales Up
TidalScale can continue to
add cores as well as memory
(and bandwidth)
Single machine solution can
only add memory

Scale, Simplify, Optimize, Evolve
Scale:
◦ Aggregates compute resources for large scale in-memory analysis
and decision support
◦ Scales like a cluster using commodity hardware at linear cost
◦ Allow customers to grow gradually as their needs develop
Simplify:
◦ Dramatically simplifies application development
◦ No need to distribute work across servers
◦ Existing applications run as a single instance, without
modification, as if on a highly flexible mainframe
Optimize:
◦ Automatic dynamic hierarchical resource optimization
Evolve:
◦ Applicable to modern and emerging microprocessors, memories,
interconnects, persistent storage & networks

Scale | Simplify | Optimize | Evolve
RESTORING DEVELOPER PRODUCTIVITY THROUGH SIMPLICTY

TidalScale Overview

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to TidalScale Overview

Similar to TidalScale Overview (20)

Recently uploaded

Recently uploaded (20)

TidalScale Overview

Editor's Notes