CMI held an event at the Fresno Grizzlie game to explore ideas of data center convergence and IBM Pure Systems, flash storage, and cloud-based backup / restore solutions from eVault
7. DC Trends
1. Modularization – Convergence
2. Virtualization
3. Cloud Computing – Public / Private / Hybrid
4. IT as a Service
5. Workload automation / management
6. Big Data meets Social - BI
7. Data Center Infrastructure Management (DCIM) -
Single Pane of Glass
8. Hadoop
9. Open Source
10. Sustainability and containing costs
7
9. Moving from Basic to Strategic
Basic data centers
Use of new technology:
43% first and fast technology
adoption
1% move virtual machines to meet
desired outcomes
21% use storage virtualization
3% use a storage service catalog
(tiered storage)
Results:
65%
New projects
Maintaining
existing
infrastructure
35%
Strategic data centers
Use of new technology:
86% first and fast technology adoption
58% move virtual machines to meet desired
outcomes
93% use storage virtualization
87% use a storage service catalog (tiered
storage)
Results
47%
53%
New projects
Maintaining
existing
infrastructure
Source: 2012 IBM Data Center Study: www.ibm.com/data-center/study ( http://www.ibm.com/data-center/study )9
14. Pure Systems: Faster time to service delivery
Expert
Integrated
System
Reduced Time, Cost and Risk
Design ManageDeploy Maintain
15. Compute
• Unmatched performance with max processor & memory capability for Intel based systems
• Full Capability Power Nodes for Unix Workloads
• Flexible choice of nodes to meet performance requirements
Network and Chassis Infrastructure
• Highest performance I/O with up to 40Gb Ethernet, 16Gb SAN speeds and/or 56Gb Infiniband FDR
• 10 Year Platform Commitment for Investment Protection
• Only platform to support a mixed environment of both x86 as well UNIX based systems
Management
• Significantly reduce the number of manual steps to greatly simplify management
• Manages IT as one; reducing complexity and automating everyday management tasks
Storage
• Reduce disk capacity needs by up to 80% with V7000 Real-time Compression
• Automatically optimize data between storage tiers with IBM Easy Tier
• Connect to ANY Major Storage Vendor – Natively or via San Volume Controller
Financial Considerations
• Significant TCO Reduction vs. Old Infrastructure
• Better Cost Basis
Pure Systems, World Class Converged Solution
16. Integrated Infrastructure
Delivering Infrastructure Services
Application Platform
Delivering Platform Services
Infrastructure Components
Beyond Blades
Data Platform
Delivering Data Services
Pure Systems – Multiple Configurations
17. PhysicalVirtualWorkload
Physical
Consolidation
& Setup
Management
Integration
Resource
Utilization
Resource
Pooling
Intelligent
Automation
Automated virtual machine placement
Dynamic allocation of virtual server, storage and
network resources*
Automated network configuration for virtual
machine deployment and mobility
Virtual workload definition
Placement services and advisors
Pooling of network switch resources to enable
consistent network policy application
Virtual Machine relocation for compute, storage
and networking
Storage resource overcommit monitoring
Space efficient copies maximize storage utilization,
accelerating deployment
Choice of
Hypervisors
Simplified Virtualization Management
19. Pure Systems Example Scenarios
• Blade Center v2.0
• DC modularization
• Disaster Recovery
• VDI
• High computational Linux platform
• “Cloud in a box”
• Rapid R&D environments
Texas Memory Systems - an IBM Company
20. Acquisition CostAcquisition Cost
SecuritySecurity
DeploymentDeployment
DowntimeDowntime
Asset ManagementAsset Management
SoftwareSoftware
MaintenanceMaintenance
PowerPower
SpaceSpace
10%-50%
50%-90%
“For modern IT platforms, 30% of the total cost
is the cost of acquiring the equipment. The
balance is for IT labor/services to configure,
maintain, upgrade, reconfigure, and ultimately
decommission the equipment.
“IT organizations are spending
from 70-80% of their total IT
budgets on maintenance and
ongoing operations.
-- Forrester-- IDC
TCO
Pure Systems – Business Case
21. Original Environment
The analysis has been performed to consolidate:
Hardware Stack
•Total of 100 servers from the Current Environment
•Total of 200 sockets and 800 cores
•Total Purchase Cost: $298,900
Software Stack
• Primary Operating System: Microsoft Windows;
License Count: 80
• Primary Virtualization Software: Microsoft Hyper-V ;
License Count: 100
• Primary Database Server Software: None ; License
Count: 0
• Primary Application Server Software: None ;
License Count: 0
IBM PureFlex Environment
The TCO and ROI analysis proposes:
Hardware Stack
• 41 IBM Flex System x240 (E5-2670) 2.6GHz
(2ch x 8co) servers
• With a total of 82 sockets and 656 cores
• Total Purchase Cost: $316,934
Software Stack
• Operating System: Microsoft Windows;
License Count: 41
• Virtualization Software: Microsoft Hyper-V ;
License Count: 82
• Database Server Software: None
• Application Server Software: None
Project Overview
23. Modularization Benefits
Consolidate
More efficiently
consolidate
systems and
applications to reduce
operating expenses
Optimize
Better tune and automate systems and
applications to improve application
performance, scalability and reliability
Accelerate Cloud
Launch self-service
applications efficiently in
a secure, and integrated
cloud environment
Innovate
More rapidly deliver new applications
and services to
meet new business needs
Consolidate Accelerate
Cloud
InnovateOptimize
23
25. The Goals
(What if you could…)
• Radically accelerate your most critical applications and stop
getting crushed at peak usage?
• Do it quickly with low risk?
• Co-exist with, and return performance and value to your
existing disc arrays?
• Derive business value from your big data in real or near real
time (Meaningful, timely analytics)?
• Minimize risk of critical production stability on early stage
products and companies?
26. Why Are There Issues in the Data Center?
In the last 10 years:
• CPU Speed: Performance increase roughly 8-10x
• DRAM Speed: Performance increase roughly 7-9x
• Network Speed: Performance increase of 100x
• Bus Speed: Performance increased roughly 20x
• Disk speed: Performance increased ONLY 1.2x
27. Price/performance ($/GB) & TCO profile favors HDDs
Used for offloading „cooler‟ data from flash storage
Optimizing for large active capacities and growth rates
„Good enough‟ performance is all you need (low I/Os
per GB)
When performance can be measured in Milliseconds
vs. Microseconds
Flash-optimized storage isn’t always the answer.
Spinning disks make sense when…
Does Spinning Disc Still Have Its Place?
28. The truth is……
• We are at a crossroad moment in data storage.
• The opportunity exists with flash technology to enable
organizations to fully utilize the power of the applications
that have been I/O bound by the constraints of
mechanical devices in an electronic data path.
• Flash is a game changer
30. Value for Storage Administrators
Simple, 2-Step Provisioning Process
– Define capacity
– Define access
Simple Active/Active Controllers
– No load-balancing required
– No complex architecture to learn
Intuitive User Interfaces
– Cloud-friendly, scriptable HTTPS API
Best-in-Class Support
– Available 24x7 to same staff as 8-5
– Support beyond the issue
Easy to Use
Great Support
Fits In Your
Environment
31. Value for Application Owners
Human Cost Recovery
– Greater productivity for application users
– Developers focus on value-add rather than tactical, iterative performance
tuning
Deterministic Performance
– Low latency all the time
Massive Scalability
– Increased performance through application parallelism
– Linear improvement without visible impact to latency
Reusable Through Upgrade Cycles
– Far ahead of CPU curve
– Upgrade Apps/Servers/Platforms and remain CPU bound
32. Flash Market Sweet Spots
Do More, Do it Faster!
OLTP Databases
– Financial, gaming, real-time billing, trading, real-time
monitoring, query acceleration (DB2/Oracle), etc.
Analytical applications (OLAP)
– Business intelligence, batch processing, ERP
systems, reporting, massive data feeds, etc.
Virtual Infrastructures
– VDI, Consolidated virtual infrastructures, user
profiles, etc.
HPC/Computational Applications
– Simulation, modeling, rendering, FS metadata, scratch
space, video on demand, thread efficiency, etc.
Cloud-scale Infrastructures
– On-demand computing, content
distribution, web, caching, metadata, GPFS, active file
management, etc.
33. Combining Flash with Storage Virtualization
Clients gain…
• Extreme performance of IBM FlashSystem with IBM
MicroLatency™
• Advanced storage functionality of IBM SVC
• Thin Provisioning – allocate storage “just in time”
• Easy Tier – storage efficiency
• FlashCopy – point in time copies
• Mirroring/Copy Services – data replication and protection
• Real-Time Compression – up to 5X more data in the same physical space
• An ability to cost effectively deploy quickly and realize
immediate results
34. Equinix Case Study
Equinix
– Global colocation provider with data centers in 31 markets across 15
countries in the Americas, EMEA, and Asia-Pac
Challenge
– In the middle of a major update to their “order to cash” business
processes including Siebel, Tibco, Oracle OBIEE. CMI asked to develop
a physical architecture for the logical application architecture.
– New architecture needed to be scalable, flexible and high performing to
grow with the company and easily add new workloads, responding on
demand to changing capacity requirements. Equinix wanted to optimize
their current storage environment (EMC) and make it easier to implement
and manage in the future.
34
35. Equinix Case Study
Resolution
– CMI took inventory of the existing architecture – applications, servers,
storage, network – and took the logical architecture sizing information to
map the logical to physical requirements.
– Reviewed options – e.g., Power vs. Intel, traditional disk vs. Flash-based
arrays, etc. and developed TCO cases to determine the best options for
compute, storage, and enterprise licensing.
• (7) IBM Power p750 servers
– Upgrades to existing p750s for reuse and cost savings
– Maximum performance, reliability, scalability and flexibility with lowest 3-year TCO
– Decreases server count to reduce Oracle license and maintenance costs
• IBM SAN Volume Controller (SVC)
– Controls the existing storage subsystems and allows for virtualization, ease of
implementation and management going forward
• Flash Memory Arrays
– Supports the new data needs for the rollout – approx. 120TB
– Improves performance 50-80% over traditional disk systems to reduce I/O
constraints
– Lowers TCO – fewer servers/storage needed as a result of the performance gains
35
36. Equinix Case Study
Impact
– Equinix rolled out the new infrastructure and is delighted with the
performance, reliability, flexibility, ease-of-management and overall value
of the entire solution
– Team was trained quickly on the new platform and have confidence in
their ability to manage and scale the solution as the business requires
– Improved performance of databases by 60-80%
– Reduced Oracle licensing and maintenance costs by 25%
– Reduced hardware costs by 23%
– Simplified ongoing storage provisioning and management
36
38. Cloud Backup Types
Vendor Managed
Vendor managed backup
solution. Client specified
requirements, installs the
agents / or uses client
software. The vendor
provides a managed
service.
Public
Backups
Shared Management
Client managed cloud
backup destination. A
replacement for existing
backup hardware.
Private
Backups
Pub
Backups
Array Solution
Client managed cloud
storage appliance that
provides centralized
management and local
access to data.
SitePvt
Appliance
38
39. Cloud Vendor Managed Backup
Vendor Managed
Private
Backups
Charge Model:
Charge based on data transferred. This is an inexpensive solution if
your data change rate is low.
Best suited for:
Companies wanting SLA‟s and to get out of the backup management
business.
Setup:
Client discovery is done to specify SLA‟s, storage tiers, archive and
compliance requirements as well as identify constraints like bandwidth
and backup windows.
Backup Process:
Typically vendor agents are installed on hosts and cloud hardware is
dedicated.
Restore Process:
Self service web portal as well as 24/7 customer service.
Typical Options:
Local appliance, de-duplication, compression and network
acceleration, proactive reporting.
39
40. Cloud Shared Management Backup
Public
Backups
Shared Management
Charge Model:
Based on data capacity stored in the cloud.
Best suited for:
Organizations that have a high change rate with a lower amount of
data. A replacement for offsite backup hardware, while retaining full
control of backup management and software.
Setup:
Storage is allocated and the initial copy is made over the wire. Some
providers offer a one time seed option to speed the initial backup.
Backup Process:
Vendor agent or host mapped drive. Managed by client owned backup
software.
Restore Process:
Client owned backup software or self service web portal.
Typical Options:
Management, 24/7 customer service, local appliance, de-duplication,
compression and network acceleration.
40
41. Cloud Array Backup
Pub
Backups
Array Solution
SitePvt
Appliance
Charge Model:
One time local appliance hardware charge, plus identified cloud
storage providers.
Best suited for:
Organizations wanting centralized management of multiple cloud
storage solutions, while retaining full control of the backup
management and software. Typically includes: de-duplication,
compression and network acceleration.
Setup:
Installation of appliance and configuration of cloud providers.
Backup Process:
Vendor agent or host mapped drive. Managed by client owned backup
software.
Restore Process:
Local appliance storage cache or cloud self service web portal.
Typical Options:
HA, DR, Site to Site replication
41
42. Cloud Considerations
• Total capacity stored and data change rate
• Backup Management and SLA
• Licensing Costs
• Staff Costs. HR issues. Training issues.
• Equipment. Floor/Rack space
• Network Bandwidth
• Tiering Options: reduces cost of storage. 50 TB of spinning tier one disks?
• Compression / De-duplication Option: reduces amount of data stored
• Optional Functionality: Backup, HA, DR.
– It is your backup. By default it‟s HA. By default it speaks top DR.
• CAPEX and OPEX spend
• Restoring data is not usually as simple as one tape = one load.
• Adds discipline, SLAs, structure to Service Levels
• Focus on production/applications
42
43. Operational Recovery Time
43
• Does not include transfer time
• Assumes tapes are located onsite and available
• Iron Mountain or off-site = more time
• On-site? How often checked, tested?
• What if the label is wrong?
Step Tape Minutes Cloud Minutes
1 Go to PC 5 Go to PC 5
2 Browser 5 Browser 5
3 Find File 5 Find File 5
4 Find tape (Assumes tape is onsite) 60 Restore File End
5 Load Tape 30
6 File Span multi tapes 30
7 Restore File End
Example Recovery Time* 2.25 hrs .25 hrs
* Does not include transfer time
44. Data Recovery Time
Assumptions
• WAN Speed - 50Mb dedicated up and down
• LAN Speed - 100Mb dedicated
• 50 GB file restore example
• 100 MB file restore example
• Tape restore is performed over the network
Restore time calculated by: data quantity converted to megabits divided by the transfer rate. 8 bits = 1 byte.
• 100MB = 800Mb / transfer rate = transfer time in seconds.
• 50GB = 50000MB = 400,000Mb / transfer rate = transfer time in seconds.
44
Backup Technology
100 MB Restore
Time
50 GB Restore
Time Process Delivery
Recovery
SLA
Tape 7.6 seconds 63 minutes Manual Local Tape
Client
Managed
Cloud Managed Backup 15.25 seconds 127 minutes
Live Person or Web
Portal Restore Point Yes
Cloud Vendor Managed
Backup 15.25 seconds 127 minutes Web Portal Web Portal Yes
Cloud Vendor Managed
Backup with appliance 7.6 seconds 63 minutes Web Portal
Local
Appliance Yes
Cloud Array Backup 7.6 seconds 63 minutes Online
Local
Appliance Yes
45. Cloud Backup Vendors
Example Customer Spend Use Case
• 50TB of mixed data (80% files, 20% database)
• 50Mb Pipe (shared)
• Multiple OS/Hosts (78 Windows / 19 AIX)
45
Vendor (Parent
Company)
Data
Managed
Example Yearly
Spend
Spend
Type Term
Client
Staffed
Management
SLA
Data
SLA Experience Other items to note:
Tape
Adic Scalar LTO4 50TB CAPEX
Depreciation
Cycle Yes No No. Client Staff
- Hardware is dedicated to client, not
shared
Cloud Managed Backup
IBM Managed Cloud
Backup (IBM) 50TB $540K + $50K setup OPEX
Typically 3 year
contracts, many
options No Yes
Five
9's
Enterprise solution
10 years
140 PB+ a month
- Hardware is dedicated to client, not
shared
- Recommends that the circuit needs
to be upgraded to an estimated
108.2Mbps.
eVault (Segate)
30TB
Compresse
d $600K OPEX
12 month
contract only Yes No
Five
9's
Enterprise solution
15 years
100 PB+
35k customers
40% cloud market
- Private cloud option.
- Rebranded by many (Fujitsu,
SunGard).
Cloud Vendor Managed
Backup
Nirvanix (IBM) 50TB $80.4K OPEX
Typically 1 Year,
many options Yes No
Five
9's
Enterprise solution
5 years
40 PB+ stored
- Requires you to use your own
backup software.
Amazon S3 (Amazon) 50TB $82.3K OPEX Monthly Yes No
Four
9's
Enterprise solution
6 years
- Requires you to use your own
backup software.
eVault (Segate) 50TB
$96K + $300K
software
CAPEX/OPE
X
12 month
contract only No Yes
Five
9's
Enterprise solution
15 years
100 PB+
35k customers
40% cloud market
- Private cloud option.
- Rebranded by many (Fujitsu,
SunGard).
Cloud Array Backup
Twin Strata (Twin
Strata) Unlimited $30K + Cloud Provider
CAPEX/OPE
X Support Contract Yes No
Not
Applic
able 2 years
Whitewater (Riverbed) Unlimited $30K + Cloud Provider
CAPEX/OPE
X Support Contract Yes No
Not
Applic
able 2 years
46. EVault Back up Deployment Options
EVault Cloud
(SaaS)
Client Systems
On Customer Site
EVault Data Transfer Process
Deduplication, Compression,
and Encryption
EVault Software or Appliances
On Customer Site
EVault Cloud
(2nd Vault)
2nd
Customer Site
EVault
CentralControl
Agents Data Transfer Vault Replication
I want to talk a few minutes about combing flash with storage virtualization. What is storage virtualization? – Let’s pause for a moment and go back to 2005, just 8 years ago. That is when the push for server virtualization really started and chances are you have Vmware deployed in your data center today and will not go back to 100% physical systems. Maybe you have KVM or Hyper-V and not Vmware, regardless. Show of hands for those that will go back to 100% physical system environment? I suspected that I would have few to no takers on that, so why are you managing your storage like you managed your servers in 2005? Now you start to understand at a high level what storage virtualization is. It breaks the bond between your servers and storage, creating a hypervisor, such that you can manage more storage arrays collectively rather than individually. Just like your Vmware Hypervisor, our storage hypervisor is intelligent. We provide all of the advanced functionality such as thin provisioning, snapshots, mirroring, automated tiering and compression for all of your storage. You no longer have to manage each array and all of the software functions on them, let the storage hypervisor do it for you. When you combine this capability with flash, you gain the ability to effectively control costs, deliver better performance across all of your storage and provide better data protection. I would like to walk your through a case study where a client took advantage of flash and storage virtualization to deliver big cost savings while accelerating application performance