Mayank Mishra

On Theory of VM Placement: Anomalies in Existing Methodologies and Their Mitigation Using a Novel Vector Based Approach

In this paper, we present the methodologies used in existing literature for Virtual Machine (VM) placement, load balancing and server consolidation in a data center environment. While the methodologies may seem fine on the surface, certain drawbacks and anomalies can be uncovered when they are analyzed deeper. We point out those anomalies and draw backs in the existing literature and explain what are the root causes of such anomalies. Then we propose a novel methodology based on vector arithmetic which not only addresses those anomalies but also leads to some interesting theories and algorithms to tackle the above mentioned three functionalities required in managing resources of data centers. We believe that with a strong mathematical base, our methodology has the potential to become the foundation of future models and algorithms in this research area.

Whither Tightness of Packing? The Case for Stable VM Placement

To date, virtual machine (VM) placement has traditionally been viewed as a bin packing problem where a number of virtual machines need to be placed on a given number of physical machines. The goal of bin packing based approach is to minimize the number of physical machines (PMs) used. However, the resource utilization of VMs is dynamic unlike static artifacts in the bin packing problem and varies with the workload being handled by applications on the VM. This means that tight packing may result in a situation where the applications run out of resources needed to handle the workload since there is no room to expand. This will trigger a VM migration to another PM having sufficient resources to allow the VM to expand. Migration is an expensive process both in terms of the resources needed for migration as well as in terms of the degradation of application performance during migration. A simple solution to this is to simply provision every VM for its peak usage-however this results in wasted resources since peaks occur infrequently and only for short durations of time. This peak usage based approach is one of very loose packing which is inefficient but provides stability of the VM in the context of migration. What we need is a balance between the tightness of packing (to optimize the number of physical machines used) and VM stability (to minimize the number of the migrations resulting from a given placement). In this paper we propose a metric to quantify this notion of balance and an algorithm to place VMs so as to minimize “imbalance”. We show through extensive experiments that balanced placement is not so loose as to be inefficient while giving us good stability as measured from the number of resource shortfalls that would occur with a given placement. In our experiments we found that balanced placement results in 15 to 80 percent lesser resource shortfalls and upto 90 percent less severe resource shortfalls when compared to other placement schemes.

Low cost computing using virtualization for Remote Desktop

The goal of this project is to provide a low cost open source Remote Desktop based computing environment to users by using Virtualization Technology and existing open source software and tools. In this project, we have used LTSP (Linux Terminal Server Project) [1] to access Remote Desktop and Xen [2] hypervisor to provide virtual desktop environment at server. A user gets a personalized desktop in the form of a VM (Virtual Machine) running over remote server. It is different from the old Remote Desktop solutions in a way that instead of a login session on single OS remote server, user will get a full fledged desktop with desired OS. The most important benefit for the system will be the resource conservation as facilities like live VM migration [3] available with virtualization will help in Load Consolidation and Load Balancing. Because of virtualization, users desktop is also isolated from other users in the same physical machine. This project makes use of open source tools like LTSP and XEN to make the solution affordable to everybody. There are certain similar solutions available in market like Citrix Xen Desktop [4], VM Ware View [5] but because of their proprietary nature they are both costly and closed systems.

Bulk I/O Storage Management for Big Data Applications

We propose and design a new Block I/O schedulingscheme called Bulk I/O Dispatch (BID) suited for disk intensiveMapReduce applications. Large data access by such applicationsresult in a large number of block I/O requests which have thepotential to be sequentialized. However, due to contention byother applications and the way current I/O schedulers operate, the opportunities of a large sequential I/Os are missed. SequentialI/Os, which are faster than random I/Os, can lead to saving theCPU wait times and thus better application performance. The proposed scheduler is designed to work with all blockdevices which have superior sequential performance than random. Through simulation based experiments with MapReducebenchmarks we show that the proposed block I/O schedulerresults in about 27% to 52% lesser time for I/O than the currentlyavailable schedulers.