América Latina
2008
Modern Data Centers:
Creating a New Value Proposition
Inter-secretarial Commission
for the Development of Electronic Government Mexico City
Data Center Consolidation
•
Consolidating data centers is much more than physical movement of servers and infrastructure from many locations to fewer sites.•
The facility costs and power savings that can be achieve are just the beginning of the opportunity•
The biggest benefits come from using the consolidation initiative to−
Rationalize the application portfolio and reduce license cost−
Virtualize the server, storage, and network infrastructure−
Reduce labor and/or cost of external services associated with maintenance activities−
Archive inactive data−
Monitor the quality of active data−
Monitor overall performance to improve service deliveryLarge Data Centers
Small Data Center
Questions for You
•
How many data centers are represented in this room?
•
How many that are
−
Large >5,000 sq ft
−
Medium 1,000 – 4,999 sq ft
Classification of Data Centers
Slide 7 Redundant capacity – Redundant capacity
components but only single non-redundant electrical path distribution
Concurrently Maintainable - Redundant capacity
components and multiple electrical path
distribution but only one path serves computing equipment at any time
Fault Tolerant Site - Redundant capacity components and
multiple electrical path distribution simultaneously serving computing equipment
Basic Infrastructure - Non redundant
capacity components and single non redundant electrical path distribution
Service Availability
Area Tier I Tier II Tier III Tier IV
Operational
Implications Site is susceptible to disruption from
planned and unplanned activities and events Site may be susceptible to capacity component
failure and will be susceptible to a distribution path failure Site is susceptible to disruption from unplanned activities Site is not susceptible to a single worst-case unplanned event or any planned maintenance activity Average Annual impact of maintenance/ unplanned outages Average Downtime for a Tier I data center is estimated at 28.8 hours per year or 99.67% availability Average Downtime for a Tier II data center is estimated at 22 hours per year or 99.75%
availability
Average
Downtime for a Tier III data center is estimated at 1.6 hours per year or 99.98%
availability
Average
Downtime for a Tier IV data center is estimated at 0.8 hours per year or 99.99% availability
Attributes of Small “Data Centers”
•
Smaller facilities have limited scale and access to few resources•
Since support personnel are few, they tend to be less organized and more vulnerable to staffing absences•
Since these sites generally have less access to financial resources:−
They are less secure,−
Less able to operate within recommended temperature and humidity ranges,−
Generally unmonitored after working hours•
Small data centers of this sort are highly susceptible to disruption from planned and unplanned activities and events.−
Average downtime is at least 65 times greater than a Tier 4 data center and at least 17 times greater than a Tier 3 data centerConsolidation: An Active Trend
•
U.S. Federal Government initiative
−
Baseline ~2,900 data centers
−
Goal
1,132 data centers
•
Canada
−
Baseline 140 data centers
plus
single purpose
centers
−
Goal
10 data centers
•
Australia
−
Goal
Cost Savings of
$1 Billion AUS
•
General Motors
−
Baseline
23 major data
centers
−
Goal
2 major data
centers
•
HP
−
Baseline 85 major centers
−
Goal
6 major centers
•
Intel
−
Baseline 130 major centers
States Are Consolidating
•
At the state level within the United States, data center consolidation has been active since 2007•
Despite varying definitions for what constitutes an enterprise data center, the trend towards consolidation appears to be universal•
States have been consolidating their computing assets intoraised floor, secured, centralized data center facilities.
•
Many states are utilizing remote back-up data center facilities for the purpose of back-up anddisaster recovery and business continuity.
Why Consolidate?
•
Improve service levels
•
Reduce costs
•
Make strategic moves to or away from certain platform types
•
Reduce staff levels or avoid further hiring
•
Align systems with the enterprise's organization
•
Improve focus on business, operational excellence, or architecture issues
•
Support systems added in a merger or acquisition (Datacenter
Rationalization) – Private Sector
Next Generation Data Center - Server
•
Server Infrastructure−
Racks of servers, with each server (or sets of servers) tied to a specific application−
Every new application brings the demand for more new servers−
Typical utilization for Intel based servers ranges from 10% to 30%•
And yet spikes do occur that can affect overall application performance•
Impact of Virtualization−
Server hardware and the actual business workload get uncoupled−
Consolidation of physical servers ranges from 10 to 20 down to 1•
In effect 20 physical servers now operate as 20 virtual servers on a single machine−
Improved application performance as virtualization software is continually making adjustments to computing resource allocations−
Enhances overall disaster recovery capability since all applications are now the equivalent of files that can be easily transferred to a DR siteNext Generation Data Center - Storage
•
Hypervisor technology at the server•
Virtualization of physical storage•
Solid state memory disk•
Multiple disk technologies within a single array•
Improved performance and data protection featuresNext Generation Data Center - Network
•
Network Switches−
New switch technology allows a single switch to support the load that would have required as many as 10 older switches−
Network virtualization enabled through software-defined technologies help organizations enable such things as Global Server Load Balancing and device interrogation.−
Software defined networks allow more robust disaster recovery and business continuity environmentNext Generation Data Center – Other Technologies
•
Blade Chassis−
HP, Cisco, Dell, IBM, etc. all make more efficient and consolidated pieces of hardware−
This has also helped improve wiring and reduced energy consumption•
Power Distribution and Racks−
Controlling rack space has allowed significant reductions in the costs associated with cooling, power, and airflow•
Single Pane of Glass Management−
Tool set consolidation to reduce license cost, improve operational coverage, promote higher skills, and deliver enhanced performanceThe Virtualization Management Gap
The Dark Side of Virtualization - Complexity
•
Headwinds−
For many organizations, as the number of virtual machines increases so does the complexity of data center infrastructure management−
As the number of virtual applications continues to grow, the cost of managing and maintaining the data center continues to increase•
Tailwinds−
High performing organizations make extensive use of automation as counter-measures to these is headwinds−
By automating the provisioning and management processes, management complexity is reduced−
The allows high performing organizations of the full benefits of virtualization and partially avoid the “virtualization management gap”Create Headwind Offsets
•
To achieve superior productivity, organizations must simplify, standardize, and automate management processes−
Automate the provisioning process to remove labor content, promote velocity, and improve quality−
Free IT staff from maintenance activities so they may focus on strategic projects that increase the organization’s competitiveness−
Reduce management cost as a percentage of total data center costs−
Remove management complexity as a barrier to further virtualizationNew Source of Latency
•
Current three-tier hierarchical networks aren’t well suited for high-volume server-to-server communications.•
Server-to-server traffic is forced to traversemultiple layers of switches, and each switch adds latency to the connection
•
Implementation of flat, low-latency networks is required to accommodate volume-intensive east-west traffic flows that are delay-sensitiveThe Problem With Latency
0 1 2 3 4 5 6 7 8 9 10th 20th 30th 40th 50th 60th 70th 80th 90th 100th Pa ge V iew s P er V isi t AOL AnalysisChange in Average Page Views Per Visit
Average Page Views Per Visit Autos Average Page Views Per Visit Money Average Page Views Per Visit Shopping
The Problem with Latency
•
Increases in latency degrade user satisfaction and in some cases also reduced revenue−
Amazon: 100 ms of extra load time caused a 1% drop in sales. (Source: G. Linden, Amazon)−
Google: 500 ms of extra load time caused 20% fewer searches. (Source: M. Mayer, Google)−
Yahoo!: 400 ms of extra load time caused a 5 to 9% increase in the number of people that clicked "back" before the page even loaded.(Source: N. Sullivan, Yahoo!)
•
Decreases in latency improve user satisfaction−
Google: trimming page size by 30% resulted in 30% more map requests. (Source: M. Mayer, Google)Slide 23
Key Technical Success Factors
•
A prior high level of operating maturity prior to starting•
Strong experience standardizing configuration across development, test, and production environments•
Solid understanding of the application portfolio and its forward demand requirements•
Effective deployment of Application Lifecycle Management (ALM) tools•
Solid experience with automated provisioning of virtual machine imagesWhat is the Journey?
Slide 25
Feasibility Assessment Architecture Implementation Manage
Business Goals Applications App Grouping Applications Enterprise Success Criteria Server/OS Server Servers Optimization Target Apps Storage Storage Storage Monitor
Target Servers Network Network Network Infrastructure Target Storage Security Security Security Security
Infrastructure Infrastructure Infrastructure Labor/Services
Is the Journey Worth it?
•
The single best metric for data center labor efficiency is the ratio of system administrators to physical servers.−
Small footprint “data centers” typically achieve ratios of 10 physical servers per 1 system administrator.−
High performing data centers invest heavily in labor saving automation•
Task automation,•
Workload monitoring,•
Virtualization of servers, storage, and network•
Process monitoring, and•
Establishment of gold build configurations−
They can achieve ratios of up to 1,000 physical servers per 1 system administrator•
It is unlikely that the Federal Government can achieve this ratio, simply because the nature of the computing load has more variability than companies like Google or Facebook.Labor Productivity and Service Availability
•
Well run, large scale data centers deliver both superior availability and lower cost than their smaller footprint counterpartsSlide 27
Tier 1 Tier 3 Tier 4
Improvement in Availability n/a 17x 65x
Administrator Ratio 10:1 100 -500 :1 500 -'1000:1 Estimated # of Administrators 1,000 100 100 Average cost per Administator $ 35,000 $ 35,000 $ 35,000 Cost for System Adminitrators $ 35,000,000 $ 3,500,000 $ 3,500,000 Recurring Savings vs. Tier 1 n/a $ 31,500,000 $ 31,500,000
Performance and Productivity Comparison versus Tier 1 Data Center Assuming 10,000 servers
Non Labor Savings
•
Approximate areas of savings−
5-15% cash creation on the value of reclaimed data center assets−
4-10% savings by installing best-in class power equipmentLower Cost is Only Part of the Value Created
•
Rationalize the Applications Portfolio−
Consolidate common applications−
Retire legacy applications•
Remove inactive data from live applications−
Many applications that are still active and useful are carrying years oftransactional data that is rarely used. Compress this data on disk and restore to production systems as necessary