+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
FLASH MEMORY: THE BIG DATA from Structure:Data 2012
1. FLASH MEMORY: THE BIG DATA
SPEAKER: Scott Metzger
VP of Analytics
Violin Memory
Friday, July 27, 2012
2. Flash Memory
The Big Data
Application
Accelerant
Q1 2011
Scott Metzger, VP Analytics
Friday, July 27, 2012
3. Analytics with Big Data is
“ Good tests kill flawed
theories;
we remain alive to guess ”
again
- Karl Popper
Friday, July 27, 2012
4. Knowledge is Valuable
How much is a 10% improvement worth to you?
Friday, July 27, 2012
5. Big Data Analytics Straining
Data
Volume More
En##es
tracked Capacity
Data
richness &
Automated
data
collec#on Bandwid
th
Data
Velocity
Real-‐#me
ad-‐hoc
queries More
Con#nuous
updates IOPS
Event-‐based
triggers
Data
Variety
Lower
More
complex
queries Latency
More
fields
and
rela#onships
More
indexes
Friday, July 27, 2012
6. Memory Speed at Tier 1
1 PB
VERY FAST
No seek times
100 TB Non-volatile
Green
10 TB
1 TB
Flash Memory Arrays
8,000µs (2 orders of magnitude)
100 GB
Emulating HDDs
15K Disk Array
SATA Array
SSDs
10 GB
DRAM
Multi-core
1 GB CPU Processor Cache
ns 1µs 150µs 3ms 8ms 20ms
TIME (Access Delay)
Friday, July 27, 2012
7. Use
C ase:
F ortune
5 00
R etail
Background:
Business transactions to be warehoused every day
The longer the warehousing process takes the earlier the
warehousing process needs to start
Growing number of transactions and richness of data add to
challenge of managing data
Results:
Flash memory array performance over spinning disk akin to 1G to
4G cellular data
Solid state memory physical space 1/10th spinning disk
Low latency -> more IO -> more work per unit of time
Friday, July 27, 2012
8. Use
C ase:
S ocial
N etworking
Background:
User registration rates multiplied by quarter
Exponential growth of concurrent user sessions
Additional features require more complex real-time transactions
Results:
Memory arrays external to servers meet high availability
requirements so when a component fails apps are still running
Consistent user experience; page load time, custom content serving
80% reduction in power and cooling requirements
Less compute requires less license cost
Friday, July 27, 2012
9. Use
C ase:
U .S.
G overnment
Background:
Cost pressures for consolidation and desktop virtualization
Virtual Desktop Infrastructure (VDI) is difficult to scale
Training teams waiting too long for desktop and application boot
times
Results:
1/10th the wait time
‘Non-virtualized’ user experience for desktop apps
Regained confidence in viability of VDI can be scaled
Friday, July 27, 2012