Since now all I've read about cloud elasticity implementation was about adding/removing VMs to/from the cloud server. But is it possible to gain elasticity without adding more VMs? In other words, is it possible to enlarge the resources of existing VM in cloud server in an automatic manner?
For example, assume we have a VM with one cpu core-1.2GHz and 4GB Ram and the cloud management system detects the VM would require more resources. So to achieve elasticity, we must add one more cpu core and 4GB ram to the VM. But in papers I read that it can be done by adding more VM to meet this requirement. In fact instead of adding resources to the existing VM, they add an other VM to support user demands.
Yes, you certainly can, using VMWare's Distributed Resource Scheduler and/or Resource Pools. If you assign all the machines in a cluster or similar to the same resource pool, and allocate more than the needed CPU/RAM, DRS allows VirtualCenter to automatically allocate extra resources as needed by resource rather than a host-by-host level of granularity. See this research paper on the topic, as it provides a bit more detail than I can quickly type up:
http://golftreeinternational.com/other/VMWare/module%2011/page%20519/vmware_drs_wp.pdf
Hi Rasul,
I will set my answer first to a theoretical framework. It is interesting to define that there are two types of adaptability (i.e. horizontal and vertical). Therefore there are two ways to add more resources, either by scaling up (down) or scaling out (in). It is interesting to research on these definitions.
Now in terms of adding more resources to an existing VM, concerning Openstack ,this update cannot be performed on the fly, but you have to stop the VM, to modify the flavor settings and then to start it again. Therefore, the service must be interrupted.
Now if you want to add a more powerful VM, then you should consider to instantiate this VM in parallel with the first one, and then to steer the traffic from the first one to the second in a seamless way. Depending on the type of the service, such an action is possible, state tracking issues may be raised.
Hi Rasul,
Cloud elasticity depends upon the physical resources that a machine has. In order to modify the Guest OS (In terms of adding resources to it ) It requires a refreshing mechanism to tell the System of the cloud that RAM or CPU has been increased. So a refreshing mechanism is required always.
A CPU used predefined clock cycle for its speed and System bus to Access the RAM . If you are changing those parameters a system level interrupt always require.
On the other hand if we think that the Cloud provider controls the management of the physical resources like amount of RAM /CPU assigned to Vm then with little interrupt it can control this thing by considering the concept of Plug and play devices.
Regards,
Wajeeta Lohana
Of course there is alot of work in Vertical elasticity....where they increase capacity of resources in real-time.
See these papers:
http://www.diva-portal.org/smash/get/diva2:751209/FULLTEXT01.pdf
http://www.infosys.tuwien.ac.at/staff/sfarokhi/soodeh/papers/Soodeh-Farokhi_CameraReady_CAC-2015.pdf
http://ieeexplore.ieee.org/xpl/login.jsp?tp=&arnumber=7152516&url=http%3A%2F%2Fieeexplore.ieee.org%2Fxpls%2Fabs_all.jsp%3Farnumber%3D7152516
https://scholar.google.se/citations?view_op=view_citation&hl=en&user=H5ddpX8AAAAJ&citation_for_view=H5ddpX8AAAAJ:TFP_iSt0sucC
Yes, it is possible. As Harilaos Koumaras said before, we could use horizontal and vertical elasticity.
I published two papers that could help in this sense:
https://www.researchgate.net/publication/235534087_A_Survey_on_Cloud_Computing_Elasticity
https://www.researchgate.net/publication/281620475_A_Programming-Level_Approach_for_Elasticizing_Parallel_Scientific_Applications
The first one is available to download. The second I could send to you if you have interest.
Regards,
Guilherme Galante
Conference Paper A Survey on Cloud Computing Elasticity
Article A Programming-Level Approach for Elasticizing Parallel Scien...
I think, a model may evolve where the resources may be assigned some elasticity parameter and written as a constrained optimization problem. I mean, a new model, without using existing routines Openstack/CloudSim may have,
- Badges
- Science method