Who is the winner of cloud computing? Comparison of OpenStack and VMware

"Editor's note" in the cloud computing ecosystem, there are two types of users that need to use cloud computing resources: the traditional (traditional IT applications) and the emerging Cloud computing Application (Cloud-aware applications) in the Internet tide. A widely circulated metaphor abroad is: in the traditional service mode, you can imagine that the server is it's pet (Pets), give them names, carefully raised, when they are sick, you have to fix them; in the new form of the application service model, the virtual machine is seen as a farm bull (cattle), Names are usually numbered, and when they get sick, you kill them and replace them with a new cow. VMware and OpenStack's cloud computing vision, features, and features are a vivid portrayal of the war or trend. The future application architecture should be like a bull on a farm: VMware's "maintenance" and the various features that protect the virtual machines hired cloud-computing application patterns may become increasingly less important. This article is widely circulated abroad, the hot discussion constantly, the Chinese community free translation to share to everybody, welcome the active discussion.

The full text reads as follows:

In the Cloud field, the most discussed is the comparison between VMware and OpenStack. In fact, it's a hot topic for people who want to use OpenStack. I have discussed this topic several times at the SF Bay OpenStack Conference, and many of my friends in the meeting want me to write these discussions down. To give readers a better understanding of the content of this article, I decided to complete this article by comparing the key points in the data center application of the two cloud computing products, which may include: open and open software systems, enterprise-class legacy applications and cloud computing applications, free software and licensing software, Through rigorous functional testing products and open function products.

The content of this article includes the following parts: Design, feature set, customer use case and value, each point is very evaluation, eventually we will be the total score to determine the winner.

The first round: design

The VMware Software Suite is a bottom-up architecture with the lower boundary as virtual Machine Manager. Like VMware's vsphere and Vcloud director products are dependent on the free ESX (i) Virtual Machine Manager, ESX (i) Virtual Machine Manager provides them with a very good deployment architecture. VMware's software suite is also fully tested and has a single deployment framework. Overall, VMware's products are used in multiple data center-scale environments because of the robustness of their architectures. In other words, VMware's software systems are closed, and the software development route is fully compliant with VMware's own development goals, and users or consumers have no control over this.

OpenStack as an open-source system, there is no single company that controls OpenStack's development path. OpenStack is young and less than three years old, but he has enormous market power, while many big companies are supporting OpenStack development (see: OpenStack supporters). With so many companies ' resources invested, the development of OpenStack is diversified. The problem, however, is that the implementation and maintenance costs of OpenStack deployments and architectures hired VMware, while technical support documents cannot keep up with the product as a result of relatively fast version updates.

VMware has a slight advantage in design, stemming from its excellent documentation and easy-to-use deployment and management interfaces. OpenStack is also in pursuit of this aspect, and it maintains its own flexibility in hardware and virtual machine management, and provides multi-vendor support.

Second round: function

VMware vMotion

Vmotion is a collection of the three main functions of vsphere DRS, DPM, and host maintenance. Where virtual machine dynamic migration allows a virtual machine to be migrated from one host to another with 0 shutdown, which is originally required for shared storage, but in vsphere 5.1, VMware does not need to implement dynamic migration through shared storage. When a virtual machine is migrated from one host to another, the memory state and data of the virtual machine are migrated synchronously. In the case of shared storage, data does not need to be migrated, only the link to the data store is changed. This speeds up the migration and reduces the load on the network during replication.

OpenStack Dynamic Migration

KVM Dynamic migration allows one virtual machine to migrate from one virtual Machine Manager to another, to be more specific, you can move a virtual machine to and from the AMD architecture host to the Intel schema master, but it is important to note that a 64-bit virtual host can only be migrated to a 64-bit host. But 32-bit has 32-bit and 64-bit options. Virtual machines cannot be manipulated during dynamic migration, but users within virtual machines can continue to work within the virtual machine. KVM is largely dependent on shared storage, which, in some ways, requires some funding.

Dynamic Migration requirements:

Virtual machine storage needs to be placed on the distributed file system, such as NFS or in Glusterfslibvirt must be open listen flag each compute node (VM Manager) must be in the same network/ The authentication between nodes in the subnet must be completed ahead of time configuring DFS mount nodes must be consistent at each compute node

OpenStack Block Storage Migration

In OpenStack, KVM supports block storage migration, which means that virtual machine migration does not necessarily require the support of shared storage. In the case of block migration, the memory state and data of the virtual machine will be migrated, but the migration operation also needs to consume the CPU resources at both ends and the operation takes time to hired the shared storage. In some user scenarios, applying block storage migrations is a good solution if we are more focused on the maintainability of the host and don't want to spend too much money. At the same time, if in the absence of shared storage environment, we want to the compute node for kernel maintenance, security upgrades, so that the virtual machine services are not interrupted, block storage migration is also ideal choice.

User scenario:

Users do not have a distributed file system, possibly due to enterprise financial support or network latency, but want to achieve high availability of virtual machines.

VMware DRS and DPM

Based on Vmotion,drs, it can dynamically monitor the current usage status of virtual machine and host, and provide support for host load balancing.

User scenario:

Deployment phase: Can perform custom automated script monitoring phase for monitoring virtual machines: DRS can distribute virtual machines on ESX (i) hosts and continuously monitor active virtual machines and available resources to dynamically migrate virtual machines to maximize resource utilization

Based on Vmotion, DPM migrates virtual machines from low load host hosts and shuts down to reduce power loss. As the load grows, DPM reboots the host and deploys new virtual machines to meet the load requirements.

OpenStack Scheduler

The OpenStack includes a scheduler for compute and volume, with a series of administrator-set rule parameters and filters that OpenStack the scheduler to deploy the virtual machine to the appropriate host. In the filter, the scheduler is very flexible, the user can complete the JSON format of the filter, and the filter also contains a lot of predefined filters. Although the OpenStack Scheduler is flexible, it is not a complete replacement for DRS for the following reasons:

VMware High Availability (highly available)

In vsphere, high availability at the virtual machine level allows the same virtual machines to be deployed on different host hosts when a virtual machine or ESX (i) host fails. There is no confusion with the fault tolerance (FT) mechanism, and the high availability is that when something goes wrong, it can be repaired within a certain amount of time. High availability is to ensure the normal operation of the virtual machine when there is a problem with the hardware, if there is a mistake, you can only start the virtual machine on the different ESX (i) host, which may cause the interruption of the service.

OpenStack High Availability (highly available)

There is no official statement of OpenStack support for high availability at the virtual machine level, which was presented in the Folsom version, but was later discarded. At present OpenStack has a hatching project evacuate, which is to provide OpenStack with virtual machine-level high availability support.

VMware Fault tolerance (fault tolerant)

The VMware fault-tolerant mechanism is to synchronize these changes to the second backup ESX (i) server by monitoring the state and all changes of the virtual machines. The concept of fault tolerance lies in the fact that the virtual machines on the host are working correctly, regardless of whether the problem occurs in the host or the hosts. Aside from the hype in marketing, this mechanism is still unable to solve the application crash in the virtual machine, because once one side crashes, the change is synchronized to the node, and when you stop the virtual machine service to fix it, the same stops the service from the node. So this mechanism can only ensure that the problem of single point failure no longer occurs, and the real application level of fault tolerance needs MSCS or WCS to solve. Considering other aspects such as maximum resource usage, memory, hard disk, CPU, bandwidth fault tolerance, these aspects have limitations and are relatively small use of the function. This requires double memory, as memory cannot be replicated across hosts, and CPU lockstepping is required to synchronize each CPU instruction. This will result in a single virtual CPU being monitored by fault-tolerant mechanisms.

OpenStack Fault tolerance (fault-tolerant)

There are no fault-tolerant features in the OpenStack, and there is no plan to complete these functions up to now. In the future, KVM also no longer supports mirroring operations.

We can see that there is still a gap between OpenStack and VMware in terms of feature support and feature details, but this is an advantage for OpenStack, because hired VMware's expensive price, OpenStack free, open advantage. VMware high input features, most of the OpenStack can be provided free of charge to customers.

VMware's lead in functionality shows that VMware continues to develop new features other than vmotion, high availability and fault tolerance to protect their virtual machines; OpenStack on the one hand, to follow VMware's footsteps, On the other hand, they put their energy into supporting more hardware vendor solutions.

Third round: Use cases

Before we evaluate the value of these functions, we first need to consider the use case problem. In the cloud computing ecosystem, there are two types of users that need to use cloud computing resources: Traditional and cloud-computing applications. Cloud-computing users will handle the HA and Dr policies themselves, while traditional users will rely on the HA and Dr provided by the cloud platform. Look at the following chart from the VMware Cloud computing Architecture article.

Common features of cloud computing applications

Distributed stateless and soft state failure switching at application end extensibility at application end

Common features of traditional application

Client-server architecture is difficult to extend the failover on the service side extensibility at the service side

Legacy applications will require features such as FT, VM-level high-availability, automated virus scanning, and cloud computing applications that do not need to be replaced by a new virtual machine when a virtual machine goes wrong.

Pet vs. cattle

To think about it in a different way, you can see the relationship between OpenStack and VMware from Microsoft's William Baker's famous article Pets vs. cattle.

Metaphor says: In traditional service mode, you can imagine your host is your pet, you give them names, such as Dusty, CERN and so on, they were carefully raised. When they get sick, you have to fix them. In the cloud-computing application service model, virtual machines are seen as bulls on the farm, their names are usually numbered, cows and cows are similar, and when they get sick, you kill them and replace them with a new cow.

The future cloud application architecture should be like a bull on a farm. VMware's maintenance, protection of the various functions of the virtual machine hired cloud computing application mode is becoming less and less important.

In this round, OpenStack after, although VMware has many OpenStack features, but for cloud computing applications, these features become less important. In the future, you're likely to pay for the uncontrollable VMware additions that you can't use.

Fourth round: Value

Now is the final round, we will decide the result of the game. Although OpenStack is more valuable than VMware, there is no clear answer to the question, and the answer depends on the size of the deployment. Although OpenStack is free, he needs a lot of engineering resources and domain experts, and he needs a lot of architecture and build work because it supports a lot of deployment scenarios, and the installation process is different. VMware will need to spend some money to purchase permissions, and relatively easier to install and run, the other hired command line, VMware learning costs lower.

Generally speaking, OpenStack entry threshold is higher, but as the scale of the project expands, you will benefit from it because there is no need to pay high copyright fees. VMware is relatively easy to install on a small scale, but as the scale expands, things change. That is to say, as the cloud application is large-scale, we are also more familiar with OpenStack, then the entry threshold of OpenStack is much lower.

In the field of cloud computing, VMware, the two-bit heavyweight, has a slightly more functional and architectural lead, but OpenStack as a underdog, catching up in the third round and giving the other a devastating blow in the final round.

PostScript

Coincidentally, at the time of writing, VMware's share price fell 22% on January 29, and the market analysis said VMware lacked a clear and excellent cloud-computing strategy.

I also understand why everyone disagrees with my points and why I have the same weights in the four-wheel comparison. To be honest, the score is not so perfect, and not so objective, but he has the meaning of his existence, he let this cloud computing this war become more interesting, please comment actively and put forward your point of view.

Translator added: For the essence of this article review

OpenStack Community: Toby Ford

This is an excellent article to dig into the difference between the two, such as Pets vs. cattle, and I think the evaluation criteria should add a few more latitude.

In the DRS vs. OS Scheduler, DRS currently has an advantage over OpenStack Scheduler because DRS uses a variety of key metrics to make decisions about host node selection for deploying virtual machines, and DRS monitors the entire lifecycle of a virtual machine.

However, DRS is closed, and these weights cannot be configured, for example: If the CPU's load suddenly increases at a very short time in the evening, this does not mean that we need to migrate the virtual machines to another host. Or it would be very difficult if the administrator knew that there would be some problems with the virtual machine in the next few years and did not want DRS to intervene. Conversely, OpenStack scheduler gradually distance itself from DRS, especially when it becomes more extensible.

It is important to explain why vMotion uses dynamic/full lifecycle to maintain virtual machines: Vmotion/drs/ha are essential for dealing with traditional virtual machines, which is obviously not related to the class of virtual machines, and I'm talking about Vmotion/drs It is important to maximize the use of resources.

In our actual environment, I had to shut down DRS because of the need to customize the scheduling rules, although we customized the scheduling rules, but the VMware upgrades made this custom scheduler very difficult to maintain.

What I want to say is that OpenStack is not only for cattle-mode scenarios, but also for virtual machines that handle pets mode.

Original link: http://www.mirantis.com/blog/cloud-prizefight-vmware-vs-openstack/

"Edit Recommendation"

Cloud computing staged "Three Kingdoms Kill" interview Oracle CEO Hurd: Cloud Computing will become the main business growth point of the cloud computing era, how to avoid critical business data from disaster attack? Red Hat launched the OpenStack focus on cloud service innovation "editor: Dan TEL: (010) 68476606"