Browse Prior Art Database

A Framework for Data Center Placement Optimization

IP.com Disclosure Number: IPCOM000239697D
Publication Date: 2014-Nov-26

Publishing Venue

The IP.com Prior Art Database

Abstract

• One of the fundamental design choice in cloud is placing VMs on the physical infrastructure • Multiple objectives to keep in mind (varies across customers) – Ensure servers/storage/network is optimally utilized – Ensure application performance is met – Ensure server-level/storage-level/network-level HA is met • Customized placement algorithms are designed for each objective • No clear way on how to combine strengths of different placement algorithms

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 37% of the total text.

Page 01 of 10

A Framework for Data Center Placement Optimization

The Problem


• One of the fundamental design choice in cloud is placing VMs on the physical infrastructure

• Multiple objectives to keep in mind (varies across customers)

- Ensure servers/storage/network is optimally utilized - Ensure application performance is met
- Ensure server-level/storage-level/network-level HA is met

  • Customized placement algorithms are designed for each objective • No clear way on how to combine strengths of different placement algorithms Methods

• Decompose the placement problem into sub-problems • Simple decomposition by server/storage/network:
- Finding a host for a VM is different from finding a storage volume for the VM (though possibly correlated)

• Hierarchical Decomposition:
- Finding a suitable rack is a separate problem from finding a host (higher-level selection restricts solution space for lower-level)


- Finding which switches a VM should use is different from which host a VM should be placed on

• Compose solutions using restricted world-views
- Output from one placement algorithms leads to a restricted world-view for other placement algorithms

• Iterate and Converge
- Allows world-views to be refined leading to acceptable solution across multiple

  criterion
- Allows multiple placement algorithms at same granularity to also be combined and achieve different objectives

1


Page 02 of 10

High-Level Flow Example

What does it need?


• A standard way to represent the different choices present for placement drives the decomposition

• A standard interface for placement algorithms allows pluggable placement algorithms
• A standard representation for data center physical topology Allows composition of

solutions

Outlining the Components

• Virtual Resource Template
- Enumeration of entities needed to describe a workload - objectives and

restrictions (e.g., workload has 3 VMs, VM1 needs 4 cores, 2 GB RAM, rack-level HA between VM1 and VM2 required)

• Physical Topology Template
- Enumeration of physical entities and their relationships. Needed for decomposition and composition.

• Physical Resource Template
- Semantic knowledge about each physical entity (e.g., this server has a SSD

2


Page 03 of 10

disk). Needed by placement algorithms.

  • Information Discovery - Physical Server Information, Storage information, network information • Placement Interface and Algorithms • Placement Execution Modules - Host Placement Adapter, Storage allocation adapter, network routing and reservation
Unified Placement Architecture

Proposed Implementation

3


Page 04 of 10

Placement Orchestrator

4


Page 05 of 10

VRT Representation


• Workload consists of a set of virtual server instances. Each instance has a virtual machine and one or more virtual volumes

- VM Attributes • RAM, Cores, Storage(DAS), Volumes
• Platform Preference (SSD etc)

- Volume Attributes
• Size, Replication Factor, Type(R/O)

• VMs can be grouped into VMGroups, Volumes can be groupe...