Вы находитесь на странице: 1из 5

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/286733848

Comparing openstack and VMware

Conference Paper · October 2014


DOI: 10.1109/ICAECC.2014.7002392

CITATIONS READS
6 2,624

2 authors:

Shalmali Suhas Sahasrabudhe Shilpa Sonawani

3 PUBLICATIONS   10 CITATIONS   
Maharashtra Institute of Technology
8 PUBLICATIONS   12 CITATIONS   
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Shilpa Sonawani on 22 August 2016.

The user has requested enhancement of the downloaded file.


 !!"#$%!##%!&'

Comparing Openstack and VMware

Shalmali Suhas Sahasrabudhe Shilpa S. Sonawani


Computer Engineering department. Computer Engineering department.
MAEER’S MIT, Pune MAEER’S MIT, Pune
Pune, India Pune, India

Abstract— Cloud computing platform relies on principle of


sharing of resources. VMware products and OpenStack are the
cloud platforms being used extensively in the industry. Of which
OpenStack is the new competitor, which provides support for
hybrid cloud environment. We compare features of VMware
products and OpenStack in this paper. Resources in cloud
include computing resources, storage and network. These
resources should be allocated and scheduled in a way that user
should get the better performance as well as provider should
achieve high resource utilizations. In this paper, we describe
VMware Distribute Resource Scheduler and OpenStack nova-
scheduler.

Keywords—cloud computing, resource scheduling, VMware


DRS, OpenStack.
Fig.1. Loosely coupled architecture of OpenStack.[12]
I. INTRODUCTION • Dashboard aka Horizon provides easy access to
Cloud computing relies on sharing of resources. compute, storage and network resources required by
Infrastructure as a Service is a type of cloud computing which virtual machines. Users can provision and relinquish
provides infrastructure as an on-demand service. IT resources from horizon.
infrastructure basically includes servers, storage, network and
operating systems. • Compute aka Nova avails computing services for
creating deleting and managing virtual machines.
OpenStack is a cloud computing platform. OpenStack
provides an Infrastructure as a Service (IaaS) and constitutes of • Network aka Neutron provides virtual networking
resources such as compute, storage and network resources.
services to Nova. It allows users to create tenant
OpenStack is a manager of multiple hypervisors. It is a
specific networks.
collection of tools for managing and orchestrating cloud
resources. VMWare ESXi is a hypervisor but not a Cloud
platform or toolkit. The VMware products that most directly • Image aka Glance can store the actual virtual disk
map to OpenStack are not vSphere or ESXi, but vCloud files in the Object Store aka swift
Automation Center and vCloud Director. In fact, OpenStack
does not have its own hypervisor but manages different • Identity management service aka Keystone
hypervisors, such as KVM, Xen, Hyper-V, AND ESXi. authenticates and authorizes all components of the
OpenStack.
In this paper we will survey design, features and scheduling
in OpenStack and VMware ESXi. VMware software suite is a bottom-up structure, the lower
boundary for the Virtual Machine Manager. Like VMware's
VSphere and vCloud director products are dependent on the
II. LITERATURE SURVEY
free ESXi Virtual Machine Manager. VMware software suite
A. OpenStack VMWare ESXi Overview itself is also fully tested, and has a single deployment
framework. Overall, VMware products because of the
OpenStack is an open source system. It is fully distributed robustness of its structure, many users have used a high- scale
system. OpenStack keeps its services as decoupled as possible.
data centers in multiple environments. The vCenter Server can
This is designed to provide massive scalability.
run database on a separate server; however it doesnot support
distributed architecture. This limits the scalability of vSphere.

 

 
Manages workload DRS features but
‘intelligently’ by Can be
grouping virtual implemented using
machines, pooling external
resources monitoring,
of physical hosts FilterScheduler and
and prioritizing Live
VMs in these Migration
pools;

Table.1. Comparison of features of VMware and OpenStack

III. VMWARE DRS AND OPENSTACK SCHEDULER


OVERVIEW

Fig.2. Tightly coupled architecture of VMware ESXi [3] Resource Scheduling in cloud is to allocate and schedule
computing resources in a way that providers achieve high
utilization of resources like compute, storage and network and
OpenStack is a Cloud computing platform or toolkit that users meet their performance requirements. This chapter gives
orchestrates various technologies such as storage, networking description of VMware Distributed Resource Scheduler (DRS)
and compute, including different hypervisors like KVM, Esxi, and OpenStack Nova-Scheduler.
Hyper-v, Xen etc. ESXi is a hypervisor but not a Cloud VMware DRS dynamically balances load as well as
platform or toolkit. The VMware products that can directly migrates VMs on increasing load to some other hosts.
map to OpenStack is not vSphere or ESXi, but vCloud
Automation Center and vCloud Director. The basic approach Resource allocation in OpenStack mainly deals with
or philosophy of the two is different [9]. Following table shows allocating compute, storage, and network resources to the
comparison of technical and non-technical features of VMware customers instance (Virtual Machine). In OpenStack, nova-
and OpenStack: scheduler makes the important allocation decisions. It currently
works with the Filter-Scheduler and a default
compute_fill_first cost function. The compute_fill_first cost
function calculates host cost based on unallocated host
Feature VMware OpenStack
memory.
Hypervisor Type 1 Type 2
virtualization, virtualization, for
A. VMware DRS
ESXi example - KVM,
ESXi, Hyper-v, VMware® DRS dynamically balances computing capacity
Xen, Baremetal across a collection of resources and intelligently allocates
Customer Access Windows Client, Open API, available resources among the virtual machines based on pre-
defined rules that reflect business needs and changing priorities
Web console, API Command line,
[8].
Horizon
(Dashboard) On initial placement, DRS makes the assumption that
Network Switching Switching, when a virtual machine being provisioned will use 100% of the
network, NSX for pluggable CPU and memory resources. It does this to ensure that it can
SDN extensions to SDN select a machine with enough resources. If DRS is ascertain
like OVS that the virtual machine cannot be hosted on selected host, it
Storage SAN, iSCSI Pluggable Cinder will make room by using vMotion and moving virtual
Image Catalogs, VM Glance image machines to another more suitable host.
Management templates and service, custom When a virtual machine experiences increased load,
OVF, upload iso to flavors and images VMware DRS first evaluates its priority against the established
Data store can be created resource allocation rules and policies, and if justified, allocates
Costs licensed Open-source additional resources [13]. Resources are allocated to the virtual
Management vCenter cannot Nova – controller machine by migrating it to another server. The live migration
system distribute its can distribute its of virtual machines to different physical servers is executed
services. services. completely transparent to end-users. Live migration is
Scheduler DRS with load Nova-scheduler, implemented by VMotion. VMware DRS can operate in two
balancing features, does not support modes viz. automatic or manual mode. In automatic mode,
VMware DRS makes decision based on best possible
distribution of VMs among the available physical resources. In
manual mode, recommendations about placements of VMs are
given and system administrator takes the final decision DRS Nova-scheduler
Resource pools can be easily added, removed or
Manages workload Uses availability zone to
reorganized. If desired, resource pools can be isolated between
‘intelligently’ by grouping group the virtual machines
different business units.
virtual machines
B. OpenStack Nova- Scheduler Does pooling of physical Not supported
Nova Compute uses the nova-scheduler service which hosts and prioritizes VMs in
selects the compute node to launch a VM. The nova-scheduler, these pools and moves VMs
like vSphere DRS, makes this automatic initial placement for load balancing.
decision using metrics, such as available compute resources. Supports affinity rules SameHostFilter,
Nova-scheduler cannot do load-balancing tasks or power- governing placement DifferentHostFilter,
management of VMs. Also nova-scheduler makes decisions of virtual machines onto GroupAntiAffinityFilter
based on static information about compute resources. It does physical hosts. are used to support
not consider current utilizations of compute and network affinity and anti-affinity
resources. The scheduler has a number of configurable options rules
and can be modified in the nova.conf file. Collects usage metrics from Ceilometer can be used to
physical hosts collect the actual
and VMs; generates utilizations of compute
optimization and network resource.
recommendations executed Currently decisions are
manually or taken based on static data
automatically – on initial for initial placement.
placement or Dynamic placement is not
continuously during runtime. supported.
VMware VMotion executes Can be implemented using
live migration of virtual external monitoring,
machines transparently. FilterScheduler and Live
Migration.
Can select particular host to Cannot select particular
provision a VM. host to launch an instance.

Fig.3. Filter Scheduler [11] Table.2. Comparison of VMware DRS and Nova-scheduler
IV. CASE STUDY
In OpenStack we have following schedulers available: To make OpenStack-IaaS enterprise ready some basic service
• Chance Scheduler: This scheduler randomly selects the levels are to be considered. Out of them availability,
host from list of filtered hosts. performance and security are most important. Availability is
mainly measured by fault tolerance (FT), high availability
• Filter Scheduler: The process of filter scheduler is
(HA) and disaster recovery (DR). Fault tolerance is important
divided in two phases. First is filtering phase. In
filtering phase lists of hosts suitable for satisfying the to guarantee availability and reliability of critical services as
request is generated. In weighing phase, hosts will be well as application execution. Fault tolerance guarantees
ranked weighted cost scores given by applying cost highest level of service availability and data protection. High
functions. Fig.3. shows the general idea of filter Availability (HA) is about enabling your systems to protect
scheduler. against and automatically recover from failures. HA has
business and application continuance. DR is defined as the
Table 2 shows comparison of the two schedulers used by "must have" coverage when the HA configuration is
VMware and OpenStack. From the table it can be seen that completely unavailable (natural disaster or any unusual
OpenStack scheduler is still lacks features like load-balancing,
events).
host selection by user, dynamic placement policy etc.
Performance can be measured in terms of applications
response times, transaction latency. Security is the major
concern for IT-applications.
The compute scheduler is responsible for determining on
which compute nodes the instance will be placed based on the
scheduler algorithm and strategies used by the algorithm.
Placement strategies are used for achieving availability, both VMware and OpenStack. DRS can take care of
performance and security levels. automatically placing VMs on a cluster and performing load
Following two are the examples for implementing placement balancing across hosts using live migration of VMs (a.k.a.
strategies. vMotion). OpenStack scheduler needs an improvement in this
part. With some customizations in the current scheduler
A. Placement Strategies: algorithm and Radware’s alteon solution OpenStack can be
• Availability- Anti affinity : used as an alternative to VMware vCloud.
The GroupAntiAffinityFilter ensures that each
instance in a group is on a different host. Application VI. ACKNOWLEDGEMENT
VMs should be placed in different ‘failure domains’ We would like to thank Department of Computer
to ensure application fault tolerance. That is even if Engineering, MAEER’S MIT,Pune, for their encouragement,
one of the application VM faces failure, VM in other extended help and suggestions.
domain remains available. This can be achieved by
making server groups and implementing affinity and
VII. REFERENCES
anti-affinity policies to these groups.
• Performance :
VMs should be placed as closely as possible on the [1] Gregor Laszewski, Javier Diaz, Fugang Wang, Geo_rey C.
network, so as to minimize latency and improve Fox. “Comparison of Multiple Cloud Frameworks.”
performance. Vol 978-0-7695-4755-8 2012 IEEE
• Security : [2] Mahjoub, M., Mdhaffar, A., Halima, R.B., Jmaiel, M. “A
Resource isolation can be done for security using Comparative Study of the Current Cloud Computing
server groups along with affinity and anti-affinity Technologies and Offers” Network Cloud Computing and
rules. Applications (NCCA), 2011 IEEE
[3] Padala, P. “Resource Management in VMware Powered
Cloud: Concepts and Techniques”, Parallel & Distributed
B. Radware’s Alteon LBaaS for OpenStack Neutron Processing (IPDPS), 2013 IEEE
[4] Abbadi, I.M., Anbang Ruan, “Towards Trustworthy
Radware’s Alteon provides an LBaaS (Load balancer as a Resource Scheduling in Clouds” Information Forensics
Service) solution natively built into the OpenStack Neutron and Security, IEEE Transactions on (Volume:8 , Issue: 6
LBaaS driver in the Havana version. Alteon significantly ), June 2013
expands open source based load balancing functionality to [5] Gupta, A, Kale, L.V, Milojicic, D., Faraboschi, P., “HPC-
offer advanced application delivery services such as SSL Aware VM Placement in Infrastructure Clouds”, Cloud
offloading, URI-based traffic steering, L7 policies and Engineering (IC2E), 2013 IEEE International Conference,
advanced health checks. Radware’s unique solution is based vol. 978-1-4673-6473-7.
on dedicated, per-tenant virtual appliance instances ensuring [6] Zhenzhong Zhang, Haiyan Wang, Limin Xiao, Li Ruan,
that tenants are not impacted by each other and that each has “A statistical based resource allocation scheme in cloud”
its dedicated capacity and networking resources. Cloud and Service Computing (CSC), 2011 International
Due to its advanced application delivery services, Alteon Conference Vol 978-1-4577-1636-2 2011 IEEE
LBaaS for OpenStack provides the following features: [8] http://www.vmware.com/in/products/vsphere/features-drs-
• Per-tenant fault isolation prevents failures of other dpm
services. [7] http://www.virtualizedgeek.com/2013/04/understanding-
• Global server load balancing allows running business vmware-vcloud-vs-openstack/
critical applications across cloud and non-cloud [8] www.rackspace.com/blog/openstack-nova-scheduler-and-
infrastructures without any additional software or vsphere-drs/
hardware. [9] http://www.mirantis.com/why-mirantis/openstack-
• Advanced health monitoring allows custom technology/comparing- vmware-and-openstack/
definition of fine grained failure scenario [10] http://networkstatic.net/comparing-vmware-vs-openstack/
• Alteon solution helps position the OpenStack solution as [11] http://docs.openstack.org/trunk/config-
a viable alternative to VMware vCloud. By enabling the reference/content/section_compute-scheduler.html
OpenStack infrastructure to offer high SLA hosting, the [12] http://docs.openstack.org
cloud becomes a viable compute platform for business- [13] http://winadmins.wordpress.com/2012/01/17/vmware-drs-
critical applications. introduction/
[14] http://www.radware.com/
V. CONCLUSION [15]https://www.ibm.com/developerworks/community/blogs/ibmsys
Cloud Computing is a powerful and flexible software sw/entry/introducing_enterprise_grade_resource_management_t
environment. Comparison between VMware vCloud and o_address_openstack_s_goldilocks_dilemma?lang=en
OpenStack is given in this paper. Also paper briefs about the
comparison between powerful resource scheduling systems of

View publication stats

Вам также может понравиться