Surety is performing system maintenance this weekend. Electronic date stamps on new Prior Art Database disclosures may be delayed.
Browse Prior Art Database

Dynamic identifiation and pinning of workloads to certain physical servers to minimize placement entropy

IP.com Disclosure Number: IPCOM000235058D
Publication Date: 2014-Feb-26
Document File: 8 page(s) / 217K

Publishing Venue

The IP.com Prior Art Database


Migration or relocation of Workloads in a DataCenter is a computation, resource and time intensive process. The approach presented in this article aims at automatically identifying workloads, based on their behavior, that trigger the computation of placement plan and hence relocation of workloads. Once identified, such workloads are isolated from the regular placement optimization process and dealt with separately, thus reducing the entropy associated with the workload placement in the datacenter. Also, disclosed is a generic approach to tag such workload so as to give hints to the placement engine about entropy causing workloads for subsequent deployment or placement of such workloads.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 21% of the total text.

Page 01 of 8

Dynamic identifiation and pinning of workloads to certain physical servers to minimize placement entropy

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 of 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 optimal utilization of servers per their  capacity. This leads to the necessity of a placement engine which does the load balancing work.  Load balancing is not an easy job and it requires consideration of many policies and finer level  details. This also leads to multitude of relocation costs as the data center tries to optimize it to  the "best balanced state". With the incoming complex policies provided by the consumers of the  data center, the placement engine logic gets complicated day by day. Add to this the erratic  nature of certain workloads which at best can be termed as "unpredictable". If the placement  engines are workload trend aware then, it becomes easier to predict future behaviors of  workloads and take a corrective action well in advance. Going by the nature of workload  distribution of workloads inside a certain aggregate of hosts or a pool of hosts - placement  engine can be fine tuned. This fine tuning shall fulfill the following goals for a data center:

1. Minimize relocation costs even when there are workloads which have very unpredictable  behavior in terms of sudden spikes or lulls.

2. Identification and segregation of such workloads into a sub zone inside the host aggregate to not  disrupt the optimal placement, of other workloads as per the policies, leading to frequent  placement or load balance calculations, which may further lead to cascading / ripple effect  throughout the Datacenter.

As a part of this idea we try to address the above two goals for the data center and propose an  apparatus that can help solve the problems mentioned above:

1. A mechanism to dynamically segregate workloads based on unpredictable behavior and pin them  to certain hosts not considered as a part of normal placement calculation inside a given pool of  hosts. Hosts refer to physical servers.

2. Associating a metadata to workload / appliance using which workload was created to identify  nature of a workload that can be directly read in order to take a pinning decision. One possible  way to achieve t...