Software Configurable Networks and Network Virtualization: Reality Check
Hello, dear ladies and gentlemen.
We nevertheless decided to diversify this blog and offer our readers not only to mark the radio button “unequivocally publish” or “fully support”, but to give our Habrovo representation an aura of controversy. We plan to take from the Internet articles on those technologies, the books on which we are in the study, translate these materials and study the reader's demand for such new developments. Below you will find a kind of trial ball: an article on virtualization and software-configured networks. We hope that the text will seem interesting, not too outdated, well translated and (sic!) Will give food for thought and comment. We will make an impression about Vox Populi and we will be very grateful to you for the contribution to our work - perhaps, so grateful that such articles will start appearing on the blog even more than once a week.
So, Jim Metzler's article on software-configured networks. Welcome under cat.
')
The technology of software-defined networks is still at the formative stage. However, the profiles of typical users of SDN networks are becoming clearer, and we can look at some of the typical evaluation criteria that such companies orient themselves to in their development. In addition, it is possible to single out such companies that seriously take up the introduction of this technology.
However, the analysis of all this information is complicated by the fact that there are many variants of SDN solutions on the market, as well as ways to use these solutions. For example, some providers focus on the dynamic movement of virtual workloads, maximizing the efficient use of the server hypervisor, in particular, technologies such as encapsulation and tunneling. Others seek to provide software network management, relying on the OpenFlow protocol to manipulate flow tables on switches.
For example, VMware and Nuage Networks can be attributed to the first camp, and NEC to the second one. The second approach strongly resembles the model promoted by the Open Network Technology Foundation (ONF), where special attention is paid to the centralization of network management. Finally, HP is in an intermediate position and seeks to bridge the gap, integrating its SDN solution into VMware tools.
Cisco stands alone. It has a number of solutions that are positioned as SDN, and some of these solutions can be attributed to the first or second SDN option described above. But Cisco’s peculiarity is that its application-oriented architecture (ACI) leaves some of the control capabilities on the switches and routers and at the same time, dedicated hardware is actively involved.
Leaving aside variations on the supplier’s side, there are four main types of SDN consumers that are emerging now. The first type is hyper-scalable capacity. So, Google designed its own SDN switches, so the whole company can work as a backbone SDN network, in which its data centers are combined. However, the capabilities of Google and most other companies, of course, are not comparable.
Another class of SDN consumer includes network service providers such as Pertino and AT & T. For example, Pertino more than a year ago launched the cloud product Network as a Service, based on SDN. In September 2013, AT & T announced the launch of the Domain 2.0 project, whose main goal is to increase the use of SDN. Considering the interest that such major players as AT & T show to SDN, as well as the presence of services like Pertino, most IT companies are likely to become familiar with SDN for the first time as consumers of such services.
The third category of potential consumers of SDN are large financial firms, such as JPMorgan and Goldman Sachs. Some of these companies are actively implementing SDN technologies, and could launch pilot projects related to SDN and other similar technologies. In March 2014, Matthew List, Managing Director of Network Core Platforms at Goldman Sachs, spoke at the Open Networking Summit conference in Santa Clara, California, optimistic about the prospects for software-defined networks, but didn’t talk about the company's plans.
It is likely that many large enterprises will indeed adopt SDN technology in the short or medium term. Considering how much attention SDNs pay to such financial corporations, there is no doubt that in 2015 we will see an active deployment of software-configured networks.
According to marketing research, all other companies do not practically use SDN yet. The author of this article had a chance to speak with a plenary report at the World Open Network Exchange (ONX) conference, where he asked the audience whose SDN companies use at least a symbolic volume. Only a few participants answered in the affirmative.
Criteria for evaluation
Given the huge interest in SDN, however, their low prevalence, we can formulate a number of criteria that allow us to give a baseline assessment of various SDN variants.
The main (and most obvious) criterion is associated with the definition of problems that we deal with in the field of IT. For example, if it is important for an IT department to support dynamic workload movement, then it is advisable to use virtualization using an overlay network. However, this approach does not allow to cope with some other problems, for example, to centralize the provision of physical routers and switches.
Other important criteria to consider are the comparison of the roles of hardware and software, as well as the degree of centralization of management. So, VMware does not hide the fact that it does not see the possibility of using specialized hardware in a data center. Cisco takes the opposite position — as mentioned above, and ACI uses specialized hardware.
Most SDN solutions use centralized network management functions. The exception is Cisco's ACI infrastructure, where some of the control functions are localized on routers and switches.
The best way that an IT company can use these two criteria to evaluate SDN solutions is as follows: simply require the supplier to justify the design decisions it has adopted. For example, is there any benefit to using specialized equipment other than increasing productivity? How does an IT organization perceive the fact that centralized management simplifies control over routers and switches, but can lead to bottlenecks?
The main criterion for evaluating any solution, in particular - a solution with centralized control - is in its scale. It is important to understand how many new tasks per second the system will be able to catch, and how this figure will change if a cluster of controllers is used instead of a single controller.
It is equally important to understand how many routers and switches allows you to use such a solution.
The following criterion attracts close attention: is the SDN solution open?
Unfortunately, the concept of "open" is very multivalued, so the answer to this question can be vague. For example, some manufacturers call their project “open” because it is based on open source.
A more common understanding of the term “open” in the context of industry standards is this: the development of open specification is conducted simultaneously by several manufacturers. A typical open standard is the OpenFlow protocol.
For all the importance of such openness, an IT company is even more in need of interoperability of its systems. Interoperability is critical, as in most cases IT companies do not receive comprehensive solutions from the same supplier. Therefore, it is necessary to confidently ensure that the various components of the solution can smoothly interact with each other.
Again, the importance of certain criteria is determined in each case, depending on the ecosystem and the goals of your company. If your primary goal is to dynamically move virtual workloads, then you probably need a network virtualization tool that uses the server hypervisor. But if you need control over all the virtual and physical devices on your network, then it would seem preferable to be such a model that takes more account of the work at the hardware level.
When?
Of course, the very first of the big financial corporations, which tried to implement SDN, have already solved the above problems for themselves, but when will other IT organizations join them? When will SDN technology reach the stage that Jeffrey Moore called the term “pragmatists” (early majority, “early majority”) in his development?
According to most sources, the abyss can be considered overcome when a new technology has taken root in 15% of companies. Another factor that shows with high confidence that this technology really justifies itself in practice is the possibility of its implementation without massive additional interventions, the absence of unpleasant surprises related to scalability and security. Finally, the chasm is likely to be overcome by the time when different IT organizations can already use the new technology as easily as more traditional technologies.
Since in the short term only a few IT organizations are going to use the SDN network without testing the concept’s performance, I asked the participants of the Network World ONX conference what percentage of IT companies, in their opinion, will launch pilot projects for working with SDN. Approximately a quarter of respondents answered in the affirmative.
Since the participants of this conference are working in companies whose interest in SDN is clearly higher than that of the average IT organization, the figure of 25% is too high. In some companies, the SDN will not pass the performance test, the priorities of these companies will change.
Taking into account all these factors, it can be assumed that in 2015, about 10% of IT organizations will be engaged in testing the performance of SDN networks. Of course, these companies will not immediately switch to full-scale industrial use of SDN technology, many will be limited to trial versions.
After a survey on the SDN health check, the author asked some of the conference participants how likely it is that the SDN technology will be involved to some extent in their organizations in the next three years. About half of the respondents believe that in a three-year term, SDNs will be used either as local area networks of their data centers, or in regional networks (WAN). Many believe that their company will use WAN services based on SDN technology, or that SDN technology will be used in campus-type corporate networks.
Since virtualization solutions based on the use of overlay networks are easier to implement than the SDN networks of the two other classes (overlay networks do not depend on the underlying physical network device), the author believes that overlay SDN networks will bridge the chasm in 2016. Two other classes of SDN networks must complete this transition at the end of 2016 or in 2017.
We nevertheless decided to diversify this blog and offer our readers not only to mark the radio button “unequivocally publish” or “fully support”, but to give our Habrovo representation an aura of controversy. We plan to take from the Internet articles on those technologies, the books on which we are in the study, translate these materials and study the reader's demand for such new developments. Below you will find a kind of trial ball: an article on virtualization and software-configured networks. We hope that the text will seem interesting, not too outdated, well translated and (sic!) Will give food for thought and comment. We will make an impression about Vox Populi and we will be very grateful to you for the contribution to our work - perhaps, so grateful that such articles will start appearing on the blog even more than once a week.
So, Jim Metzler's article on software-configured networks. Welcome under cat.
')
The technology of software-defined networks is still at the formative stage. However, the profiles of typical users of SDN networks are becoming clearer, and we can look at some of the typical evaluation criteria that such companies orient themselves to in their development. In addition, it is possible to single out such companies that seriously take up the introduction of this technology.
However, the analysis of all this information is complicated by the fact that there are many variants of SDN solutions on the market, as well as ways to use these solutions. For example, some providers focus on the dynamic movement of virtual workloads, maximizing the efficient use of the server hypervisor, in particular, technologies such as encapsulation and tunneling. Others seek to provide software network management, relying on the OpenFlow protocol to manipulate flow tables on switches.
For example, VMware and Nuage Networks can be attributed to the first camp, and NEC to the second one. The second approach strongly resembles the model promoted by the Open Network Technology Foundation (ONF), where special attention is paid to the centralization of network management. Finally, HP is in an intermediate position and seeks to bridge the gap, integrating its SDN solution into VMware tools.
Cisco stands alone. It has a number of solutions that are positioned as SDN, and some of these solutions can be attributed to the first or second SDN option described above. But Cisco’s peculiarity is that its application-oriented architecture (ACI) leaves some of the control capabilities on the switches and routers and at the same time, dedicated hardware is actively involved.
Leaving aside variations on the supplier’s side, there are four main types of SDN consumers that are emerging now. The first type is hyper-scalable capacity. So, Google designed its own SDN switches, so the whole company can work as a backbone SDN network, in which its data centers are combined. However, the capabilities of Google and most other companies, of course, are not comparable.
Another class of SDN consumer includes network service providers such as Pertino and AT & T. For example, Pertino more than a year ago launched the cloud product Network as a Service, based on SDN. In September 2013, AT & T announced the launch of the Domain 2.0 project, whose main goal is to increase the use of SDN. Considering the interest that such major players as AT & T show to SDN, as well as the presence of services like Pertino, most IT companies are likely to become familiar with SDN for the first time as consumers of such services.
The third category of potential consumers of SDN are large financial firms, such as JPMorgan and Goldman Sachs. Some of these companies are actively implementing SDN technologies, and could launch pilot projects related to SDN and other similar technologies. In March 2014, Matthew List, Managing Director of Network Core Platforms at Goldman Sachs, spoke at the Open Networking Summit conference in Santa Clara, California, optimistic about the prospects for software-defined networks, but didn’t talk about the company's plans.
It is likely that many large enterprises will indeed adopt SDN technology in the short or medium term. Considering how much attention SDNs pay to such financial corporations, there is no doubt that in 2015 we will see an active deployment of software-configured networks.
According to marketing research, all other companies do not practically use SDN yet. The author of this article had a chance to speak with a plenary report at the World Open Network Exchange (ONX) conference, where he asked the audience whose SDN companies use at least a symbolic volume. Only a few participants answered in the affirmative.
Criteria for evaluation
Given the huge interest in SDN, however, their low prevalence, we can formulate a number of criteria that allow us to give a baseline assessment of various SDN variants.
The main (and most obvious) criterion is associated with the definition of problems that we deal with in the field of IT. For example, if it is important for an IT department to support dynamic workload movement, then it is advisable to use virtualization using an overlay network. However, this approach does not allow to cope with some other problems, for example, to centralize the provision of physical routers and switches.
Other important criteria to consider are the comparison of the roles of hardware and software, as well as the degree of centralization of management. So, VMware does not hide the fact that it does not see the possibility of using specialized hardware in a data center. Cisco takes the opposite position — as mentioned above, and ACI uses specialized hardware.
Most SDN solutions use centralized network management functions. The exception is Cisco's ACI infrastructure, where some of the control functions are localized on routers and switches.
The best way that an IT company can use these two criteria to evaluate SDN solutions is as follows: simply require the supplier to justify the design decisions it has adopted. For example, is there any benefit to using specialized equipment other than increasing productivity? How does an IT organization perceive the fact that centralized management simplifies control over routers and switches, but can lead to bottlenecks?
The main criterion for evaluating any solution, in particular - a solution with centralized control - is in its scale. It is important to understand how many new tasks per second the system will be able to catch, and how this figure will change if a cluster of controllers is used instead of a single controller.
It is equally important to understand how many routers and switches allows you to use such a solution.
The following criterion attracts close attention: is the SDN solution open?
Unfortunately, the concept of "open" is very multivalued, so the answer to this question can be vague. For example, some manufacturers call their project “open” because it is based on open source.
A more common understanding of the term “open” in the context of industry standards is this: the development of open specification is conducted simultaneously by several manufacturers. A typical open standard is the OpenFlow protocol.
For all the importance of such openness, an IT company is even more in need of interoperability of its systems. Interoperability is critical, as in most cases IT companies do not receive comprehensive solutions from the same supplier. Therefore, it is necessary to confidently ensure that the various components of the solution can smoothly interact with each other.
Again, the importance of certain criteria is determined in each case, depending on the ecosystem and the goals of your company. If your primary goal is to dynamically move virtual workloads, then you probably need a network virtualization tool that uses the server hypervisor. But if you need control over all the virtual and physical devices on your network, then it would seem preferable to be such a model that takes more account of the work at the hardware level.
When?
Of course, the very first of the big financial corporations, which tried to implement SDN, have already solved the above problems for themselves, but when will other IT organizations join them? When will SDN technology reach the stage that Jeffrey Moore called the term “pragmatists” (early majority, “early majority”) in his development?
According to most sources, the abyss can be considered overcome when a new technology has taken root in 15% of companies. Another factor that shows with high confidence that this technology really justifies itself in practice is the possibility of its implementation without massive additional interventions, the absence of unpleasant surprises related to scalability and security. Finally, the chasm is likely to be overcome by the time when different IT organizations can already use the new technology as easily as more traditional technologies.
Since in the short term only a few IT organizations are going to use the SDN network without testing the concept’s performance, I asked the participants of the Network World ONX conference what percentage of IT companies, in their opinion, will launch pilot projects for working with SDN. Approximately a quarter of respondents answered in the affirmative.
Since the participants of this conference are working in companies whose interest in SDN is clearly higher than that of the average IT organization, the figure of 25% is too high. In some companies, the SDN will not pass the performance test, the priorities of these companies will change.
Taking into account all these factors, it can be assumed that in 2015, about 10% of IT organizations will be engaged in testing the performance of SDN networks. Of course, these companies will not immediately switch to full-scale industrial use of SDN technology, many will be limited to trial versions.
After a survey on the SDN health check, the author asked some of the conference participants how likely it is that the SDN technology will be involved to some extent in their organizations in the next three years. About half of the respondents believe that in a three-year term, SDNs will be used either as local area networks of their data centers, or in regional networks (WAN). Many believe that their company will use WAN services based on SDN technology, or that SDN technology will be used in campus-type corporate networks.
Since virtualization solutions based on the use of overlay networks are easier to implement than the SDN networks of the two other classes (overlay networks do not depend on the underlying physical network device), the author believes that overlay SDN networks will bridge the chasm in 2016. Two other classes of SDN networks must complete this transition at the end of 2016 or in 2017.
Source: https://habr.com/ru/post/255471/
All Articles