2.6.3
I/O virtualization
For I/O devices like the network interface card and disk, virtual interfaces are needed to be created. Multiple virtual interfaces are multiplexed onto a physical interface. Each VM has a disk partition; however, they are not allowed to access the physical disk directly. Hypervisor translates the I/O requests of the VMs to actual disk block addresses [111].
The actual mapping process depends on the hypervisor type. Type-2 hypervisor sends down the requests to host OS device drivers. The host OS treats the requests in the same way as it treats requests from other processes. However, type-1 hypervisor uses split drivers usually in conjunction with a special VM.
In the split driver technique two types of drivers are used, backend and frontend driver. The front-end driver resides in the VM and communicates with the processes of the VM. The front-end driver sends the data to the back-end driver, which resides either inside the hypervisor or on a special VM. The back-end driver communicates with the hardware to process I/O requests.
Use of virtual device offers extra security; however, at the cost of extra over- head [33]. The requests have to go through two different driver levels and translation phases; thus, adding overhead to the I/O processing. As a result, both the memory and I/O virtualization incurs more overhead compared to CPU virtualization.
All three types of resource virtualization play an essential part in Cloud operations and performance. In this dissertation, experiments have been performed with all three types of resource virtualization.
2.7
Data centers and the rise of virtual machines
In data centers and Cloud mainly type-1 hypervisor is used. A type-1 hypervisor is more efficient for simultaneously running multiple VMs on a server. In this case, the hypervisor takes full control of the hardware and use the resources efficiently. Benefits of using virtualization are many folds [5–12]. Many works have listed various benefits of virtualization; however, some of the listings are overlapping. In this dissertation the following six benefits of virtualization are identified:
2.7. DATA CENTERS AND THE RISE OF VIRTUAL MACHINES 29
2. Increase resource utilization / reducing energy cost 3. High availability of services / Fault tolerance 4. Legacy application and system
5. Maintenance
6. Pay-as-you-go model
Next, each category of the benefit of using virtualization in Cloud is discussed in brief.
2.7.1
Security / VM isolation
Security is one of the primary reasons for using virtualization in data centers [5]. A user can have administrative control over the rented VMs and install any application. However, the user does not have access to the underlying physical hardware. The data center owner can have full control of the hardware and create VMs for the users.
Virtualization allows the VM owners to use their rented VMs without compromis- ing the security of the physical server. In this way, virtualization ensures security for the data center owner without restricting the administrative privileges of the VM own- ers. The hypervisor restricts VMs from accessing the physical hardware; thus, any change made by the VM do not affect the hardware or other VMs. The hypervisor multiplexes the physical resources such as CPU, memory, and I/O in such a way that each VM is isolated from other VMs. Each VM has control over own execution en- vironment. Thus, VM owners have the full flexibility to run any application on their rented VMs.
2.7.2
Increase resource utilization / reducing energy cost
Virtualization can increase resource utilization and reduce energy cost through consol- idation [23, 23–27]. Consolidation is the process of running multiple VMs on a same physical server. Running multiple VMs on a server increases resource utilization of the physical server. In turn, that means less physical server have to be active at any given time. Thus, virtualization can help to reduce energy cost.
2.7. DATA CENTERS AND THE RISE OF VIRTUAL MACHINES 30
2.7.3
High availability of services / Fault tolerance
Virtualization helps to increase the availability of services. The key to increasing the availability of services is running multiple copies of VMs [28–30]. The high availabil- ity schemes simultaneously execute two or more copies of the same VM on different physical servers. One of the servers run the primary VM while rest of servers run the secondary VMs. The service is provided from the primary VM, while the other VMs are regularly updated with the primary VM data.
In the case, the primary VM goes down a secondary VMs is chosen to continue providing the services. The service requests are automatically diverted to the selected secondary VM. How much of the primary VM state can be recovered depends on the recovery algorithm and how many VMs are simultaneously running. Running the same task on multiple VMs in multiple locations is in the heart of the VM fault tolerance schemes.
2.7.4
Legacy application and system
There are many commercial applications still in production that was designed decades ago for the legacy hardware. Many enterprizes continue to use legacy applications and value their legacy systems highly [112–116]. All hardware has a lifetime, and they phase out eventually. Enterprises often choose to continue using the legacy applica- tions long after the life cycle of the hardware is over. When vendors stop manufactur- ing legacy hardware buying new hardware becomes impossible for the legacy software owners.
Replacing legacy hardware or application is often expensive. All the source codes need to rewritten for new compatible hardware. Furthermore, all databases and files need to be moved to the new format. Thus, moving codes and data to a new system can be costly and time-consuming. One of the cost-effective solutions is to run the appli- cation on VMs that emulate the legacy hardware. The VMs are configured to emulate legacy hardware entirely; thus it can save the cost of replacing legacy hardware.
2.7.5
Maintenance
In data centers, often a server runs for an extended period. Occasionally, the physical servers face problems and need to be shut down for maintenance. In such cases, the
2.7. DATA CENTERS AND THE RISE OF VIRTUAL MACHINES 31
VM of the host needs to be moved to another server. Live migration feature of the virtualization enables VMs to migrate while they are executing tasks [117–122].
A physical server starts showing signs of fatigue usually after long hours of opera- tion. What is more, an increase in the temperature of a part of the data center can affect the performance of the physical servers [123, 124]. The servers may also need to be shut down for other reasons like a hardware replacement or software update. In such cases, the VMs running on the servers need to be relocated to other physical servers. Live migration gives the option to relocate the VMs without interrupting the services.
The VMs can provide services to the Cloud users during the live migration pro- cess [125–127]. The actual downtime is so minimized that the VM can continue to provide services without interruption [110]. Once the migration is completed the VMs are completely moved to another part of the data center, and the old server is now free for maintenance work. Thus, using virtualization technology, it is possible to carry out maintenance tasks without interrupting the Cloud services. It is a significant advantage for the businesses.
2.7.6
Pay-as-you-go model
One of the attractive features of the public Cloud is the pay-as-you-go model for rent- ing out the VMs [128, 129]. A Cloud user can rent as many VMs as the user want and use them as long as required. Instead of acquiring hardware a user can rent VMs al- most instantly. The pay-as-you-go model is beneficial for small and medium business, which often face the surge in online user activities. VMs can be set up in minutes; on the other hand, the setting up hardware usually takes much time.
Furthermore, businesses can stop renting the VMs at any stage if the extra com- puting power is no longer required. On the other hand, it is not possible to get rid of unnecessary physical hardware easily. The model saves many start-up businesses from initially spending much money on the computing hardware. Businesses also save money by not having to pay for server maintenance and management costs.
The pay-as-you-go models are the most cost-effective option for many businesses. Thus, the virtualization played a significant role in the commercial success of the Cloud.