logging in or signing up The Mathematics of Cloud Optimization RogerSessions Download Post to : URL : Related Presentations : Share Add to Flag Embed Email Send to Blogs and Networks Add to Channel Uploaded from authorPOINT lite Insert YouTube videos in PowerPont slides with aS Desktop Copy embed code: Embed: Flash iPad Dynamic Copy Does not support media & animations Automatically changes to Flash or non-Flash embed WordPress Embed Customize Embed URL: Copy Thumbnail: Copy The presentation is successfully added In Your Favorites. Views: 1724 Category: Science & Tech.. License: All Rights Reserved Like it (0) Dislike it (0) Added: December 31, 2011 This Presentation is Public Favorites: 2 Presentation Description This Web Short describes the mathematics of cost optimization for large, mission critical cloud applications. Comments Posting comment... Premium member Presentation Transcript The Mathematics of Cloud Optimization: The Mathematics of Cloud Optimization A Web Short with Roger Sessions ObjectWatch Roger Sessions, ObjectWatch roger@objectwatch.com This presentation includes narration and is self propelled, so turn up your speakers, sit back, and enjoy the show.About Me (Roger Sessions): About Me (Roger Sessions) Author of seven books (including Simple Architectures for Complex Enterprises .) Author of dozens of white papers on IT Risk and Optimization. Fellow of the International Association of Software Architects (IASA). Multiple patents in software and enterprise architecture. Roger Sessions, ObjectWatch roger@objectwatch.comCloud Elasticity: Cloud Elasticity Roger Sessions, ObjectWatch roger@objectwatch.comCloud Elasticity: Cloud Elasticity Roger Sessions, ObjectWatch roger@objectwatch.comCloud Elasticity: Cloud Elasticity Roger Sessions, ObjectWatch roger@objectwatch.comCloud Elasticity: Cloud Elasticity Roger Sessions, ObjectWatch roger@objectwatch.comTraditional IT Cost Model: Traditional IT Cost Model Time of Day Users 0000 0400 0800 1200 1600 2000 12K 10K 8K 6K 4 K 2K Roger Sessions, ObjectWatch roger@objectwatch.com Pay for MaxCloud Cost Model: Cloud Cost Model Time of Day Users 0000 0400 0800 1200 1600 2000 12K 10K 8K 6K 4 K 2K Roger Sessions, ObjectWatch roger@objectwatch.com Pay for Actual Pay for Actual Pay for Actual Pay for Actual Pay for ActualCost to Scale: Cost to Scale Average Number of Users Cost Per Hour ($) 10 20 30 40 50 60 30 25 20 15 10 5 Roger Sessions, ObjectWatch roger@objectwatch.comCost to Scale: Cost to Scale Average Number of Users Cost Per Hour ($) 10 20 30 40 50 60 30 25 20 15 10 5 A: .50 $/user/hour Roger Sessions, ObjectWatch roger@objectwatch.comCost to Scale: Cost to Scale Average Number of Users Cost Per Hour ($) 10 20 30 40 50 60 30 25 20 15 10 5 A: .50 $/user/hour B: .05 $/user/hour Roger Sessions, ObjectWatch roger@objectwatch.comCost to Scale: Cost to Scale Average Number of Users Cost Per Hour ($) 10 20 30 40 50 60 30 25 20 15 10 5 A: .50 $/user/hour B: .05 $/user/hour What is the annual cost of running A and B with 5000 users? B = $2.1 M A = $22 M Roger Sessions, ObjectWatch roger@objectwatch.comUnits of Cloud Cost: Units of Cloud Cost Cloud Block : A Unit of Cloud Rental measured in cost/unit of time Roger Sessions, ObjectWatch roger@objectwatch.comApplications and Cloud Blocks: Applications and Cloud Blocks Roger Sessions, ObjectWatch roger@objectwatch.com Business Functions Function 1 Function 2 Function 3 Function 4 … ApplicationApplications and Cloud Blocks: Applications and Cloud Blocks Roger Sessions, ObjectWatch roger@objectwatch.com Business Functions Function 1 Function 2 Function 3 Function 4 … ApplicationCalculating Needed Cloud Blocks: Calculating Needed Cloud Blocks Roger Sessions, ObjectWatch roger@objectwatch.com Application implements B business functions. A Cloud Block can support, on average, C business functions. This application will require B/C cloud blocks.Function Clusters: Function Clusters Roger Sessions, ObjectWatch roger@objectwatch.comFunction Clusters: Function 1 Function Clusters Roger Sessions, ObjectWatch roger@objectwatch.comFunction Clusters: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Function Clusters Roger Sessions, ObjectWatch roger@objectwatch.comFunction Clusters: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Function Cluster Function Clusters Roger Sessions, ObjectWatch roger@objectwatch.comFunction Clusters: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Function Cluster Def : Two functions 1, 2 are clustered IFF a call to 1 predicts a call to 2 and visa versa. Function Clusters Let’s say F1-F10 form a cluster within the application of 100 functions Roger Sessions, ObjectWatch roger@objectwatch.comProjection: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Projection F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 Roger Sessions, ObjectWatch roger@objectwatch.com 1 Cloud Block Tight ProjectionProjection: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Projection F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 Roger Sessions, ObjectWatch roger@objectwatch.com 1 Cloud Block Say cost is $500K/year Tight ProjectionLoose Projection: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Loose Projection F1 F4 F7 F10 F13 F16 F19 F22 F25 F28 F2 F5 F8 F11 F14 F17 F20 F23 F26 F29 F3 F6 F9 F12 F15 F18 F21 F24 F27 F30 Roger Sessions, ObjectWatch roger@objectwatch.com 3 Cloud Blocks If tight projection costs $500K/year, l oose projection costs $1.5M/yearLooser Projection: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Looser Projection F1 F6 F11 F16 F21 F26 F31 F37 F41 F46 F2 F7 F12 F17 F22 F27 F32 F38 F42 F47 F5 F10 F15 F20 F25 F30 F35 F40 F45 F50 F3 F8 F13 F18 F23 F28 F33 F38 F43 F48 F4 F9 F14 F19 F24 F29 F34 F39 F44 F49 5 Cloud Blocks If tight projection costs $500K/year, looser projection costs $2.5M/yearA Unified Architectural Approach to the Cloud: A Unified Architectural Approach to the Cloud Roger Sessions, ObjectWatch roger@objectwatch.comA Unified Architectural Approach to the Cloud: A Unified Architectural Approach to the Cloud Capability Focused Capability consists of function cluster The IT systems that support those functions And the data used by those systems Roger Sessions, ObjectWatch roger@objectwatch.comA Unified Architectural Approach to the Cloud: A Unified Architectural Approach to the Cloud Capability Focused Technical connections reflect business connections Data is not shared outside the capability Capability consists of function cluster The IT systems that support those functions And the data used by those systems Cross capability connections are loosely coupled (asynchronous) Roger Sessions, ObjectWatch roger@objectwatch.comA Unified Architectural Approach to the Cloud: A Unified Architectural Approach to the Cloud Capability Focused Technical connections reflect business connections Business boundaries define technical and data boundaries Data is not shared outside the capability Capability consists of function cluster The IT systems that support those functions And the data used by those systems Cross capability connections are loosely coupled (asynchronous) Roger Sessions, ObjectWatch roger@objectwatch.comA Business/IT Partnership: A Business/IT Partnership Roger Sessions, ObjectWatch roger@objectwatch.comA Business/IT Partnership: A Business/IT Partnership Business defines business clusters IT reflects these clusters The Team owns the Capability Architecture Roger Sessions, ObjectWatch roger@objectwatch.comReview: Review The tightness of the projection of function clusters onto cloud blocks is critical to cloud cost control. The approach we use to achieve tight projection is through a capability based architecture. Because the business function clusters define the capability boundaries, we must get them right. Roger Sessions, ObjectWatch roger@objectwatch.comThree Challenges: Three Challenges 1. Combinatory Explosion 2. Sensitivity 3 . Timing Roger Sessions, ObjectWatch roger@objectwatch.comProblem 1. Combinatory Explosion: Problem 1. Combinatory Explosion This is a partitioning problem. Total Functions Number of Partitions 1 2 3 1 2 5 4 15 5 52 6 203 7 877 8 4,140 9 21,147 10 115,975 Roger Sessions, ObjectWatch roger@objectwatch.comProblem 2. Sensitivity: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Problem 2. Sensitivity 10 % error F1 F2 F3 F4 F5 F6 F7 F8 F9 F11 F10 100 % overrun Roger Sessions, ObjectWatch roger@objectwatch.comProblem 3. Timing: Problem 3. Timing We can’t determine clusters until after deployment. We need to know the clusters before we implement. Roger Sessions, ObjectWatch roger@objectwatch.comWhat is the Solution?: What is the Solution? We have a mathematical problem. Let’s look for a mathematical solution. Roger Sessions, ObjectWatch roger@objectwatch.comSIP: SIP SIP: Simple Iterative Partitions Roger Sessions, ObjectWatch roger@objectwatch.com Based on the mathematics of complexity and partitions. It drives the best possible capability architecture to deliver the tightest possible projection. It addresses all three of the difficult issues.SIP Directed Clustering: SIP Directed Clustering a b Synergistic? Yes No a b a b Roger Sessions, ObjectWatch roger@objectwatch.comSIP Directed Clustering: SIP Directed Clustering a b Synergistic? Yes No a b a b Definition of Synergistic - S S(A,B) = True, if, from the business perspective, A is not useful without B and visa versa. Note: This definition makes S an equivalence relation . Roger Sessions, ObjectWatch roger@objectwatch.comEquivalence Relations: Equivalence Relations Directed There is only one possible outcome, so mistakes are eliminated. Property Implications Roger Sessions, ObjectWatch roger@objectwatch.comEquivalence Relations: Equivalence Relations Directed There is only one possible outcome, so mistakes are eliminated. Rational The outcome can be tested for validity Property Implications Roger Sessions, ObjectWatch roger@objectwatch.comEquivalence Relations: Equivalence Relations Conservation of Structure We can partition the system before implementation. Directed There is only one possible outcome, so mistakes are eliminated. Rational The outcome can be tested for validity. Property Implications Roger Sessions, ObjectWatch roger@objectwatch.comMore Information: More Information On Equivalence Relations: On SIP: The Mathematics of IT Simplification (White Paper) by Roger Sessions at www.objectwatch.com . SIP: An Introduction (Web Short) by Roger Sessions at www.authorstream.comSummary: Summary Cloud optimization can reduce costs by a full order of magnitude. Optimization is a mathematically difficult problem. SIP is a methodology focused on large scale cloud optimization. The problem requires a mathematical solution. Roger Sessions, ObjectWatch roger@objectwatch.com Even small errors are very expensive.Summary: Summary Cloud optimization can reduce costs by a full order of magnitude. Optimization is a mathematically difficult problem. SIP is a methodology focused on large scale cloud optimization. The problem requires a mathematical solution. Roger Sessions, ObjectWatch roger@objectwatch.com Even small errors are very expensive. Interested in a SIP workshop at your company? Like a copy of my upcoming white paper Cloud Optimized Architectures? Want to hear about future Web Shorts? Drop me a note: roger@objectwatch.com Twitter: @ RSessions Linked In: Roger SessionsPowerPoint Presentation: The End You do not have the permission to view this presentation. In order to view it, please contact the author of the presentation.
The Mathematics of Cloud Optimization RogerSessions Download Post to : URL : Related Presentations : Share Add to Flag Embed Email Send to Blogs and Networks Add to Channel Uploaded from authorPOINT lite Insert YouTube videos in PowerPont slides with aS Desktop Copy embed code: Embed: Flash iPad Dynamic Copy Does not support media & animations Automatically changes to Flash or non-Flash embed WordPress Embed Customize Embed URL: Copy Thumbnail: Copy The presentation is successfully added In Your Favorites. Views: 1724 Category: Science & Tech.. License: All Rights Reserved Like it (0) Dislike it (0) Added: December 31, 2011 This Presentation is Public Favorites: 2 Presentation Description This Web Short describes the mathematics of cost optimization for large, mission critical cloud applications. Comments Posting comment... Premium member Presentation Transcript The Mathematics of Cloud Optimization: The Mathematics of Cloud Optimization A Web Short with Roger Sessions ObjectWatch Roger Sessions, ObjectWatch roger@objectwatch.com This presentation includes narration and is self propelled, so turn up your speakers, sit back, and enjoy the show.About Me (Roger Sessions): About Me (Roger Sessions) Author of seven books (including Simple Architectures for Complex Enterprises .) Author of dozens of white papers on IT Risk and Optimization. Fellow of the International Association of Software Architects (IASA). Multiple patents in software and enterprise architecture. Roger Sessions, ObjectWatch roger@objectwatch.comCloud Elasticity: Cloud Elasticity Roger Sessions, ObjectWatch roger@objectwatch.comCloud Elasticity: Cloud Elasticity Roger Sessions, ObjectWatch roger@objectwatch.comCloud Elasticity: Cloud Elasticity Roger Sessions, ObjectWatch roger@objectwatch.comCloud Elasticity: Cloud Elasticity Roger Sessions, ObjectWatch roger@objectwatch.comTraditional IT Cost Model: Traditional IT Cost Model Time of Day Users 0000 0400 0800 1200 1600 2000 12K 10K 8K 6K 4 K 2K Roger Sessions, ObjectWatch roger@objectwatch.com Pay for MaxCloud Cost Model: Cloud Cost Model Time of Day Users 0000 0400 0800 1200 1600 2000 12K 10K 8K 6K 4 K 2K Roger Sessions, ObjectWatch roger@objectwatch.com Pay for Actual Pay for Actual Pay for Actual Pay for Actual Pay for ActualCost to Scale: Cost to Scale Average Number of Users Cost Per Hour ($) 10 20 30 40 50 60 30 25 20 15 10 5 Roger Sessions, ObjectWatch roger@objectwatch.comCost to Scale: Cost to Scale Average Number of Users Cost Per Hour ($) 10 20 30 40 50 60 30 25 20 15 10 5 A: .50 $/user/hour Roger Sessions, ObjectWatch roger@objectwatch.comCost to Scale: Cost to Scale Average Number of Users Cost Per Hour ($) 10 20 30 40 50 60 30 25 20 15 10 5 A: .50 $/user/hour B: .05 $/user/hour Roger Sessions, ObjectWatch roger@objectwatch.comCost to Scale: Cost to Scale Average Number of Users Cost Per Hour ($) 10 20 30 40 50 60 30 25 20 15 10 5 A: .50 $/user/hour B: .05 $/user/hour What is the annual cost of running A and B with 5000 users? B = $2.1 M A = $22 M Roger Sessions, ObjectWatch roger@objectwatch.comUnits of Cloud Cost: Units of Cloud Cost Cloud Block : A Unit of Cloud Rental measured in cost/unit of time Roger Sessions, ObjectWatch roger@objectwatch.comApplications and Cloud Blocks: Applications and Cloud Blocks Roger Sessions, ObjectWatch roger@objectwatch.com Business Functions Function 1 Function 2 Function 3 Function 4 … ApplicationApplications and Cloud Blocks: Applications and Cloud Blocks Roger Sessions, ObjectWatch roger@objectwatch.com Business Functions Function 1 Function 2 Function 3 Function 4 … ApplicationCalculating Needed Cloud Blocks: Calculating Needed Cloud Blocks Roger Sessions, ObjectWatch roger@objectwatch.com Application implements B business functions. A Cloud Block can support, on average, C business functions. This application will require B/C cloud blocks.Function Clusters: Function Clusters Roger Sessions, ObjectWatch roger@objectwatch.comFunction Clusters: Function 1 Function Clusters Roger Sessions, ObjectWatch roger@objectwatch.comFunction Clusters: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Function Clusters Roger Sessions, ObjectWatch roger@objectwatch.comFunction Clusters: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Function Cluster Function Clusters Roger Sessions, ObjectWatch roger@objectwatch.comFunction Clusters: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Function Cluster Def : Two functions 1, 2 are clustered IFF a call to 1 predicts a call to 2 and visa versa. Function Clusters Let’s say F1-F10 form a cluster within the application of 100 functions Roger Sessions, ObjectWatch roger@objectwatch.comProjection: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Projection F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 Roger Sessions, ObjectWatch roger@objectwatch.com 1 Cloud Block Tight ProjectionProjection: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Projection F1 F2 F3 F4 F5 F6 F7 F8 F9 F10 Roger Sessions, ObjectWatch roger@objectwatch.com 1 Cloud Block Say cost is $500K/year Tight ProjectionLoose Projection: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Loose Projection F1 F4 F7 F10 F13 F16 F19 F22 F25 F28 F2 F5 F8 F11 F14 F17 F20 F23 F26 F29 F3 F6 F9 F12 F15 F18 F21 F24 F27 F30 Roger Sessions, ObjectWatch roger@objectwatch.com 3 Cloud Blocks If tight projection costs $500K/year, l oose projection costs $1.5M/yearLooser Projection: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Looser Projection F1 F6 F11 F16 F21 F26 F31 F37 F41 F46 F2 F7 F12 F17 F22 F27 F32 F38 F42 F47 F5 F10 F15 F20 F25 F30 F35 F40 F45 F50 F3 F8 F13 F18 F23 F28 F33 F38 F43 F48 F4 F9 F14 F19 F24 F29 F34 F39 F44 F49 5 Cloud Blocks If tight projection costs $500K/year, looser projection costs $2.5M/yearA Unified Architectural Approach to the Cloud: A Unified Architectural Approach to the Cloud Roger Sessions, ObjectWatch roger@objectwatch.comA Unified Architectural Approach to the Cloud: A Unified Architectural Approach to the Cloud Capability Focused Capability consists of function cluster The IT systems that support those functions And the data used by those systems Roger Sessions, ObjectWatch roger@objectwatch.comA Unified Architectural Approach to the Cloud: A Unified Architectural Approach to the Cloud Capability Focused Technical connections reflect business connections Data is not shared outside the capability Capability consists of function cluster The IT systems that support those functions And the data used by those systems Cross capability connections are loosely coupled (asynchronous) Roger Sessions, ObjectWatch roger@objectwatch.comA Unified Architectural Approach to the Cloud: A Unified Architectural Approach to the Cloud Capability Focused Technical connections reflect business connections Business boundaries define technical and data boundaries Data is not shared outside the capability Capability consists of function cluster The IT systems that support those functions And the data used by those systems Cross capability connections are loosely coupled (asynchronous) Roger Sessions, ObjectWatch roger@objectwatch.comA Business/IT Partnership: A Business/IT Partnership Roger Sessions, ObjectWatch roger@objectwatch.comA Business/IT Partnership: A Business/IT Partnership Business defines business clusters IT reflects these clusters The Team owns the Capability Architecture Roger Sessions, ObjectWatch roger@objectwatch.comReview: Review The tightness of the projection of function clusters onto cloud blocks is critical to cloud cost control. The approach we use to achieve tight projection is through a capability based architecture. Because the business function clusters define the capability boundaries, we must get them right. Roger Sessions, ObjectWatch roger@objectwatch.comThree Challenges: Three Challenges 1. Combinatory Explosion 2. Sensitivity 3 . Timing Roger Sessions, ObjectWatch roger@objectwatch.comProblem 1. Combinatory Explosion: Problem 1. Combinatory Explosion This is a partitioning problem. Total Functions Number of Partitions 1 2 3 1 2 5 4 15 5 52 6 203 7 877 8 4,140 9 21,147 10 115,975 Roger Sessions, ObjectWatch roger@objectwatch.comProblem 2. Sensitivity: Function 1 Function 2 Function 3 Function 4 Function 5 Function 6 Function 7 Function 8 Function 9 Function 10 Problem 2. Sensitivity 10 % error F1 F2 F3 F4 F5 F6 F7 F8 F9 F11 F10 100 % overrun Roger Sessions, ObjectWatch roger@objectwatch.comProblem 3. Timing: Problem 3. Timing We can’t determine clusters until after deployment. We need to know the clusters before we implement. Roger Sessions, ObjectWatch roger@objectwatch.comWhat is the Solution?: What is the Solution? We have a mathematical problem. Let’s look for a mathematical solution. Roger Sessions, ObjectWatch roger@objectwatch.comSIP: SIP SIP: Simple Iterative Partitions Roger Sessions, ObjectWatch roger@objectwatch.com Based on the mathematics of complexity and partitions. It drives the best possible capability architecture to deliver the tightest possible projection. It addresses all three of the difficult issues.SIP Directed Clustering: SIP Directed Clustering a b Synergistic? Yes No a b a b Roger Sessions, ObjectWatch roger@objectwatch.comSIP Directed Clustering: SIP Directed Clustering a b Synergistic? Yes No a b a b Definition of Synergistic - S S(A,B) = True, if, from the business perspective, A is not useful without B and visa versa. Note: This definition makes S an equivalence relation . Roger Sessions, ObjectWatch roger@objectwatch.comEquivalence Relations: Equivalence Relations Directed There is only one possible outcome, so mistakes are eliminated. Property Implications Roger Sessions, ObjectWatch roger@objectwatch.comEquivalence Relations: Equivalence Relations Directed There is only one possible outcome, so mistakes are eliminated. Rational The outcome can be tested for validity Property Implications Roger Sessions, ObjectWatch roger@objectwatch.comEquivalence Relations: Equivalence Relations Conservation of Structure We can partition the system before implementation. Directed There is only one possible outcome, so mistakes are eliminated. Rational The outcome can be tested for validity. Property Implications Roger Sessions, ObjectWatch roger@objectwatch.comMore Information: More Information On Equivalence Relations: On SIP: The Mathematics of IT Simplification (White Paper) by Roger Sessions at www.objectwatch.com . SIP: An Introduction (Web Short) by Roger Sessions at www.authorstream.comSummary: Summary Cloud optimization can reduce costs by a full order of magnitude. Optimization is a mathematically difficult problem. SIP is a methodology focused on large scale cloud optimization. The problem requires a mathematical solution. Roger Sessions, ObjectWatch roger@objectwatch.com Even small errors are very expensive.Summary: Summary Cloud optimization can reduce costs by a full order of magnitude. Optimization is a mathematically difficult problem. SIP is a methodology focused on large scale cloud optimization. The problem requires a mathematical solution. Roger Sessions, ObjectWatch roger@objectwatch.com Even small errors are very expensive. Interested in a SIP workshop at your company? Like a copy of my upcoming white paper Cloud Optimized Architectures? Want to hear about future Web Shorts? Drop me a note: roger@objectwatch.com Twitter: @ RSessions Linked In: Roger SessionsPowerPoint Presentation: The End