Dynamically increase host-capacity by reducing host margins using Work-load co-relation based on utilization data
Publication Date: 2014-Mar-28
The IP.com Prior Art Database
AbstractThere are various ways to achieve workload consolidation while providing IaaS to the customers of cloud computing. Providing stable plans for workload consolidation involves optimal usage of resources. Historical usage data of workloads give a pattern of workload behavior that can be used to predict the future behavior of a workload. This submission provides a mechanism to do workload consolidation using the historical data with the help of traditional bin packing algorithm.
Page 01 of 7
Dynamically increase host-capacity by reducing host margins using Work -load co-relation based on utilization data
A data center comprises of several physical servers and these servers can then be logically segregated based on data center preferences to host workloads. The cloud user is unaware of the hardware hosting a given workload, since the data center takes care of resource allocation for the hardware by provisioning a Virtual Machine on one of the underlying servers. The data center is also responsible for doing load balancing to ensure that none of the servers are loaded upto capacity. This leads to finding out an optimal margin per physical server for different type of resources including the CPU, Storage and Network needs. The margin for a given host can be defined as the percentage range of the total capacity of the server that should not be allotted to a Virtual Machine in order to account for variance in demands of individual workloads hosted on that server. If the margin is not chosen carefully then the Physical server may fail in hosting the workloads appropriately with the best throughput. In more generic terms, the data center would want to avoid any Predictive Failure Alerts on the hardware using a very safe margin.
We would try to dynamically assign capacity margins for physical servers to ensure maximum utilization viz fulfilling more workload demands as well as choose the right number to not throw the server in to failures. This is done by using workload corelation, which is aimed at reducing relocation costs as well. In this article we describe the following:
1. Apparatus to perform dynamic workload corelation to get a stable placement plan that reduces the cost of relocation using heuristics.
2. Use the corelation data to decrease the host margin dynamically to fulfill more workload demands and hence increasing greater hardware usage.
With the expansion of cloud and relocation costs involved with the movement of workloads across different physical servers, it becomes necessary to produce more stable placement plans, reducing the delta between placement plans. Generally workloads have patterns of usage. These usage directly affect the resource demands of the Virtual Machine they are hosted on and in turn affects the Physical Servers. Workloads can have complimentary behavior as well, and hence if put together, can subsequently compliment each other on resource demands. For example, Workload1 - may host applications that are accessed between 8 AM to 6PM and similarly Workload2 might have peaking r...