Mais conteúdo relacionado
Semelhante a Track 1, Session 2, Flash by Amit Sharma (20)
Mais de EMC Forum India (20)
Track 1, Session 2, Flash by Amit Sharma
- 1. Flash 1st – A powerful data centre
optimization and consolidation
approach
1 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 2. 2009: More Apps On Virtual Infrastructure
17,500,000
15,000,000
The Tipping Point Virtual Machines
12,500,000
Physical Hosts
10,000,000
7,500,000
5,000,000
2,500,000
2005 2006 2007 2008 2009 2010 2011 2012 2013
Source: IDC
2 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 3. Virtualization Changes Everything…
SLA optimization has shifted from Point-in-Time to Real-Time
MANUAL PROCESS 100% AUTOMATED
Oracle Exchange SQL
server server server
Oracle Exchange SQL OracleI ExchangeI SQLI
Automated Tiering
SSD SAS NL-SAS
Static Association of Dynamic, Self-optimizing
Disk Groups to Applications Storage Pools
based on best guess based on actual data activity
3 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 4. Information Tipping Point Ahead
This decade will be nothing like the last
140,000
120,000
Data
100X
100,000
A 10TB data center in 2001
80,000 growing at 60% YoY will be a
120 Petabyte data center by
60,000 MORE DATA 2021, but IT budgets remain flat
EACH DECADE Automation
40,000
20,000
Budget
0
2000 2003 2006 2009 2012 2015 2018 2021
4 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 5. Most Data Will be Stone Cold
While we increasingly can’t erase data, we can store it better
4,123
100 X
37.5
TB
TB
2011
2021
Hot Warm Cold
Evolution of 50 TBs in 10 Years
5 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 6. The CPU to HDD Performance Gap
CPU improves 100 times every decade – disk speed hasn’t
Moore’s Law will continue to improve
CPU performance while disk drive
performance will remain flat.
As a result, applications will suffer more
and more unless we rapidly move to
FLASH
10,000 times
FLASH improved
100 times
improved
2000 2010 2020
6 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 7. Clicker Question
Have you faced the following challenges.
A. Due to performance requirements your storage solution
was built of a huge number of spinning drives whereas
your capacity requirement was much lower?
B. You still prefer 15K spindles and have your arrays totally
filled up?
C. You were forced to do incremental purchases over the life
span of your infrastructure to meet performance SLA’s.?
7 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 8. Anatomy of an Enterprise FLASH Drive
Designed for reliability, data integrity and performance
Controller
SLC NAND
FLASH
SAS or SATA
ports
DRAM
End to End CRC
8 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 9. 9 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 10. Comparing Associated Costs
Which technology is the most efficient?
Capacity Acquisition Cost Capacity Power Cost
$- $10 $20 $30 $40 0 5 10 15 20 25 30
15K HDD 15K HDD
7200 HDD 7200 HDD
FLASH FLASH
FLASH 7200 HDD 15K HDD FLASH 7200 HDD 15K HDD
$/GB $34.80 $0.43 $1.63 mWatt/GB 25 6 28
Transaction Acquisition Cost Transaction Power Cost
$- $2 $4 $6 $8 $10 $12 0.0 50.0 100.0 150.0
15K HDD 15K HDD
7200 HDD 7200 HDD
FLASH FLASH
FLASH 7200 HDD 15K HDD FLASH 7200 HDD 15K HDD
$/IOPS $1.99 $9.56 $5.44 mWatt/IOPS 1.4 133.3 94.4
10 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 11. How?
11 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 12. The FLASH 1st Data Strategy
Hot data on fast FLASH SSDs—cold data on dense disks
“Hot”
high activity
Highly active data is As data ages, activity
stored on falls, triggering automatic
FLASH SSDs for movement to high capacity
fastest disk drives for lowest cost
response time
Data Activity
FLASH High Cap.
SSD HDD
Movement
Trigger “Cold”
low activity
Data Age (5 years)
12 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 13. Clicker Question
Have you had discussion with your solution suppliers around
the mix of flash and spinning drives
• Yes – we had detailed discussions and some sizing methods
were implemented
• No – we had no discussions
13 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 14. How Much FLASH?
Used dynamically with Automated tiering
Size of FLASH FIFO • Older data is constantly
being replaced by new
Amount of Data being highly active data
created daily
• The amount of FLASH
required is determined by:
– The amount of data created
FLASH Capacity
each day, and
– The period of time it takes to
cool
Number of Days of High
Data Activity
14 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 15. How Much FLASH?
3 fundamental business questions
• How much data are you storing
today?
50 TBs
• How much is your data growing
each year?
50% YoY
• How long does your data stay hot?
60 days
15 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 16. Calculating Needed Flash
Net new is a function of amount of “starting data” and “growth”
• 50% of 50TB is 25TB
• The average amount of data
50% growth rate generated each day:
80 – 25 X 1024 GB / 365 =
70 GB per day
60 25
• FLASH Capacity:
40 – 60 days X 70 GB =
50 50 4,200 GB
20
0 • FLASH Percentage:
Start Year 1 – FLASH Capacity/Total Capacity
– 4,200 GB/(75 X 1024) GB X 100:
6%
16 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 17. Clicker Question
What has been your biggest hindrance to flash 1 st strategy.
A. We feel flash is too expensive
B. We preferred continuing with spinning spindles as it was
tried and tested
C. We need help to ascertain our workloads
D. We feel with flash and advanced tiering there is
complexity.
17 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 18. Workload Skew Defines Configuration and benefits
Take Advantage of Workload Skew
Heavy Skew Moderate Skew Low Skew Zero Skew
95% of I/O on 90% of I/O on 80% of I/O on 50% of I/O on
5% of data 10% of data 20% of data 50% of data
Configuration A Configuration B Configuration C Configuration D
3% EFD 3% EFD 3% EFD
25% FC 15 K
43% FC 15 K 80% FC 15 K
97% SATA 72% SATA
54% SATA 20% SATA
30% More 40% More 20% More Same
Performance Performance Performance Performance
80% Fewer Disks 60% Fewer Disks 50% Fewer Disks 17% Fewer Disks
20% Lower Costs 15% Lower Costs Same Cost 13% Lower Costs
18 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 19. Automatic Data Optimization – FAST VP
The benefit of FLASH without the cost
• Easy and Effective Implementations
– Per-application Policies, modeled by
professional tools
– Works with everything, validated against
real workloads
Active Inactive
5% 95%
Data FLASH Disk Data • Exclusion policies
• SLA based Tiering – Promote/Demote policies
SSD HDD
• Granular Data Management
– Minimum chunk of movement
– Batch vs realtime , frequency of analysis
• Advanced Controls
– Time controls for performance and
movement
– 3 Tiers, dynamic, online
90% utilization
19 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 20. FAST Cache
• Fully automatic application accelerator
– System-wide resource, immediate impact
– No extensive pre-planning needed
FAST Cache – Helps both Read and Write operations
– Quick adjustment to changing usage patterns
– Up to 2TB
Flash
Fibre
• Turbo-charge performance improvement in
Channel Oracle, Microsoft SQL Server, and VMware View
environments
Turbo-charge performance
SATA
Benefits Easy to manage and monitor
20 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 21. Disk Utilization
Application was experiencing higher response time-Disk Utilization touching 80%
Before FAST Cache
After FAST Cache
• Application response time
improved
• Drive utilization reduced.
• Response time increased
• Additional Headroom
21 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 22. Lets change the way we procure IT
Faster Response Times
Smaller Footprint
Less Power
Lower Cost
22 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
- 23. THANK YOU
Thank You
23 Cloud Meets Big Data
© Copyright 2011 EMC Corporation. All rights reserved. 17-18 November 2011. Grand Hyatt - Mumbai
Notas do Editor
- The number of virtual machines has overtaken the number of physical server. Gap increasing further. Virtual 1st policy.