IT As Service Meichun Hsu ( 许玫君 玫君 ) HP Labs China 2006/5/8 1 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף �ِ�ِ�
Outline of Talk • Technology and Economic Trends • Selected Research at HP Labs • Concluding Remarks 2006/5/8 2 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
IT as Service Technology and Economic Trend 2006/5/8 3 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Economic Trend of IT 2006/5/8 4 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Next Generation IT Center: Consolidation and Virtualization Application Stacks (silos) Application Stacks De-couples resources from consumption Virtualization Virtual Virtual Virtual Virtual Resource Resource Resource Resource Consolidated Duplicated Infrastructure Infrastructure Without Virtualization: With Virtualization: • Duplicated, fragmented management • Consolidated management • Duplicated, under-utilized resources • Consolidated, highly-utilized resources • Vertically-integrated teams • Specialized, nimble teams • Fragmented expertise • Centers of expertise 2006/5/8 5 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
An Example IT Consolidation Case Benefits Before Now • 66% reduction in data centers 3 utilized 2 1 1 • Reduction from 4 custom and 6 payroll PeopleSoft instances to 1 single global instance of PeopleSoft 8 63% server reduction • − From 41 to 10 dedicated − From 0 to 5 utility servers Cycle time improvement – some • applications have gone from development to production in 6 weeks Almost a 4x reduction in • application infrastructure specific deployment costs 60% reduction in storage (disk • space) 7 TB 3 TB 2006/5/8 6 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
The economics of managing a data center will fundamentally change Today Consolidated Solution (fewer people, more capacity) software is primary enabler Typical IT Ratios Today: Typical IT Ratios NGDC: One person � 20 servers One person � 200 servers One Person � 2TB storage One Person � 200TB storage Source: IDC Source: HP analysis A Case Study Before New Consolidation Solution Existing Network (100 servers) New (100 virtual servers) − − 20 Infrastructure, 80 Capacity Servers 4 ProLiant DL580 servers manage 100 VMs − − 12 computer racks required 2 MSA1000 SANs host storage − − High maintenance requirements (2.5 FTE) Low maintenance requirements (1.0 FTE) Up-Front Hardware & Installation $700k Up-Front Hardware & Installation $165k Hosting ($1k/month/rack) $144k Hosting ($1k/month/rack) $ 12k H/W support contracts + $100k HW support contracts + $ 5k Maintenance FTE + $200k Maintenance FTE + $100k Annual cost = $444k Annual cost = $117k Net result of almost 75% annual cost savings, 60% labor savings, and 94% reduction in physical hw (servers/storage) 2006/5/8 7 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
The next-generation IT center enables & requires running IT as Utility Services • Provided by an infrastructure invisible to the end- user • Standard interface and properties • Expected to be always available • Low cost to plug in to • Payments aligned with usage • Delivered as a service Consolidation + Virtualization = Running IT as Utility Services 2006/5/8 8 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
IT as Utility Service - Selected Research • Capacity Planning for Server Consolidation • Data Center Automation • Market-based Resource Allocation • Technology for Support Services • Grid Computing 2006/5/8 9 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Capacity Planning for Server Consolidation Alex Zhang et al 2006/5/8 10 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Server Consolidation Problem • Problem description: − Given a set of old servers and their associated workload traces (e.g. CPU utilization time series), how can we “ pack ” them into a minimum set of new servers? Servers before consolidation Output: Number of servers required and assignment VMM VMM Servers after consolidation 2006/5/8 11 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Fitting Jigsaw Puzzles with Probabilistic Goals Probabilistic Capacity Limit t = 1 t = 2 t = 3 t = 4 • Issues CPU − Can workloads be proportionally scaled? − Are Workloads additive? t = T t = 1 t = 2 − Metrics independence? Memory t = 3 t = 4 − What is the VMM Overhead? t = T Servers Servers 2+5 Server 1 Server 2 Server 3 Server 4 Server 5 1+3+4 =New = New Server 1 Server 2 • Probabilistic goals − E.g. 5-minute CPU utilization < 50% to be satisfied with probability 0.995 2006/5/8 12 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
A High-Dimensional Bin-Packing Algorithm for Server Consolidation The problem input data is 1. The CPU utilization traces (objects) w ( i , a , t ), for server i = 1, 2, … , n , performance metric a = 1, 2, … , K , and time interval t = 1, 2, … , T ; 2. The bin capacities C ( a ) (same over all time intervals, but potentially different across performance metrics a = 1, 2, … , K ); and α (such as 0.99) for satisfying the bin capacity (potentially different across performance metrics a = 1, 2, … , K ). 3. The probability ( a ) m ∑ = (1) Minimize Z y ( j ) = • Optimization Model: j 1 Subject to: − Given: A set of old servers to be packed; ≤ = = (2) x ( i , j ) y ( j ), for i 1 , 2 ,..., n and j 1 , 2 ,..., m . bin capacity (on multiple metrics) with m ∑ = = (3) x ( i , j ) 1 , for i 1 , 2 ,... n . Probabilistic goals = j 1 n ∑ − High-Dimensional Bin-Packing Formulation ⋅ − ⋅ ≤ ( , , ) ( , ) ( , , ) ( , , ) ( ), w i a t x i j M j a t v j a t C a (4p) = i 1 = = = for j 1 , 2 ,..., m , a 1 , 2 ,..., K and t 1 , 2 ,..., T . • Achieved an implementation of complexity O(n 2 m T ) (5) x ( i , j ) = 0 or 1 (binary variable). ≤ ≤ (6) 0 y ( j ) 1 (continuou s variable) . − n is number of old servers, − m is number of metrics, ∑ T ≤ 1 − α = = (7p) v ( j , a , t ) [ ( a )] T , for j 1 , 2 ,..., m and a 1 , 2 ,..., K . − T is number of time periods (m*T is the dimensionality) = t 1 ≤ = = = (8p) v ( j , a , t ) y ( j ), for j 1 , 2 ,..., m , a 1 , 2 ,..., K and t 1 , 2 ,..., T . (9p) v ( j , a , t ) = 0 or 1 (binary variable). 2006/5/8 13 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Data Center Automation 2006/5/8 14 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Hierarchy of abstraction Datacenter service bus Datacenter service bus Application utility Application utility The future Data Virtual Virtual Virtual Virtual Services catalog system system system system (virtual resources) Center is one that services services services services offers an intelligent Mechanisms and Capacity Capacity Mapping Mapping infrastructure that planning planning workflows provides Virtual resource pools Virtual resource pools appropriate Resource pools (virtual in this example) resources, that is highly automated Server utility Storage utility Data fabric Data fabric Resource pools (physical in this example) 2006/5/8 15 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Model Based Management Automation - Quartermaster Software Architecture Quartermaster Core • Model and instance repository Quartermaster Model Manager(s) • Model creation and management • Model translation Quartermaster Tools • System composition • Capacity management • Resource Allocation • Reservation/Scheduling • … 2006/5/8 16 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Thermal Cooling dynamic data center thermal management • Power density becoming critical – affects reliability and cost • Use modeling and measurement to understand thermal characteristics of data centers − Saving 25% today • Exploit this for dynamic resource allocation and proper provisioning − Today 40 1u boxes in a rack in a data center − creates 10Kw rack − cooling in by intuition • Smart cooled data center - shift workload to cool areas or cool where the workload is • In a 15 Mw data center, we can save $1Million/year 2006/5/8 17 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Tycoon A Market-Based Resource Allocation System Kevin Lai, Lars Rasmusson, Li Zhang, Eytan Adar, and Bernardo Huberman 2006/5/8 18 וֹכּמּף ףץ٪ّ٠מּَِ ٩٭۶ףוֹ٭٩ץף ێ ۖףףף�ِ�ِ�
Recommend
More recommend
