Juniper Node Slicing and Universal Chassis
Seeing what is at the top of the stairs being somewhere in the middle is as difficult as understanding which trends in the telecommunications industry will firmly strengthen and become the foundation of future developments, and which will disappear without a trace. Not so long ago, as part of the Cloud-Grade Networking design, Juniper Networks announced the emergence of a new mechanism for virtualizing its products, and also introduced the concept of a universal chassis for routing and switching applications. News is important in terms of practical application and interesting as a reason to guess about the future.
First, marketers, and then engineers began to call modern routers multiservice. After all, it's great when a single chassis can solve routing tasks at peer-to-peer junctions, terminate broadband access subscribers, transmit transit MPLS traffic, pull packets out of the data center overlay, and even do address translation and protocols. The versatility and reduction of equipment seems reasonable and to people in suits who are good at counting Capex / Opex, and people who calculate stance and the required power of the UPS look at monitoring alerts at night, take inventory, configure or maintain this equipment. because with one entity it is easier to manage than with ten. Everything would be fine, but as the number of functions increases, internal contradictions appear between them. Imagine a multi-appliance that washes, prepares lunch and stores winter tires, because of imperfect software, lunch can turn out not tasty, because the ingredients accidentally got a little laundry detergent, and the tires can be deformed due to the close proximity of the oven. In my practice, I worked with clients, routers, which during the life cycle turned into a monolith. It is no longer possible to disassemble it, because “this is our everything”, and it becomes increasingly difficult to exploit, because one function, again due to the imperfect software, hinders the implementation of the other.
Our IT colleagues have long since resolved similar problems for themselves - server administrators are increasingly using virtualization and containers, and developers are actively decomposing entities and writing micro-services. Until recently, network engineers could only afford virtualization, as one friend from Odessa said - virtualization for beggars, in the form of separate routing tables, management contexts or logical routers. This adds stability and convenience since it shares tasks at least at the level of the processes of a network operating system, but does not solve the problem of solidity globally. And finally, Donyel Jones-Williams Director of Service Provider Product Management from Juniper Networks stated the following - We’ve been doing this for hypervisors. In my opinion, it turned out even more interesting, because Junos Node Slicing, the so-called Juniper Networks virtualization system, involves running instances of the operating system not on RE, but on separate servers, this is how it looks.
')
As you can see from the illustration, GNF (guest network function) is an instance of Junos, runs as a separate virtual machine and exclusively receives the set of resources placed on FPC slots. Each GNF with an FPC set is seen from the outside and functions as a completely separate router, with traffic between GNF can be sent through the factory. This approach eliminates internal contradictions between the functions of the router, increases the stability of the system as a whole, and besides it has the “natural” ability to divide administrative management domains between organizational units. Allow one more illustration, they began to draw them very nicely.
The soon appearance of a universal chassis, where interface cards from PTX10008, PTX10016, QFX10008, QFX10016 and MX can be used, and this is what is reported in a press release from Juniper Networks, causes only one question - And what was it possible? Juniper already has some developments in the field of unification, for example, some of the MX240-MX960 line card hubs can be used on the top MX2010 and MX2020 for a long time. The current task looks more serious, since in one chassis it is necessary to combine the maps designed taking into account the peculiarities of work in all network segments - data center, core, boundary. How this is done is not reported yet, but according to Donyel Jones-Williams, the new platform is already in test operation in the networks of large customers.
So what's the future of trying to bring Juniper Networks closer? The manufacturer concentrates not only on the equipment, but mixes the accents in the direction of control and flexibility of operation. According to various estimates, the cost of the final product is only 40% determined by the hardware, about the same is the cost of software development, integration and automation tools. Let's see what happens and, if possible, we will take part in the development of the industry, because no one knows where the staircase will lead, as Doc Brown said - the future has not yet been written.
The press release and documentation on Junos Node Slicing can be viewed here:
→ Juniper Networks Unveils Cloud Grade Networking
→ Junos Node Slicing Feature Guide
First, marketers, and then engineers began to call modern routers multiservice. After all, it's great when a single chassis can solve routing tasks at peer-to-peer junctions, terminate broadband access subscribers, transmit transit MPLS traffic, pull packets out of the data center overlay, and even do address translation and protocols. The versatility and reduction of equipment seems reasonable and to people in suits who are good at counting Capex / Opex, and people who calculate stance and the required power of the UPS look at monitoring alerts at night, take inventory, configure or maintain this equipment. because with one entity it is easier to manage than with ten. Everything would be fine, but as the number of functions increases, internal contradictions appear between them. Imagine a multi-appliance that washes, prepares lunch and stores winter tires, because of imperfect software, lunch can turn out not tasty, because the ingredients accidentally got a little laundry detergent, and the tires can be deformed due to the close proximity of the oven. In my practice, I worked with clients, routers, which during the life cycle turned into a monolith. It is no longer possible to disassemble it, because “this is our everything”, and it becomes increasingly difficult to exploit, because one function, again due to the imperfect software, hinders the implementation of the other.
Our IT colleagues have long since resolved similar problems for themselves - server administrators are increasingly using virtualization and containers, and developers are actively decomposing entities and writing micro-services. Until recently, network engineers could only afford virtualization, as one friend from Odessa said - virtualization for beggars, in the form of separate routing tables, management contexts or logical routers. This adds stability and convenience since it shares tasks at least at the level of the processes of a network operating system, but does not solve the problem of solidity globally. And finally, Donyel Jones-Williams Director of Service Provider Product Management from Juniper Networks stated the following - We’ve been doing this for hypervisors. In my opinion, it turned out even more interesting, because Junos Node Slicing, the so-called Juniper Networks virtualization system, involves running instances of the operating system not on RE, but on separate servers, this is how it looks.
')
As you can see from the illustration, GNF (guest network function) is an instance of Junos, runs as a separate virtual machine and exclusively receives the set of resources placed on FPC slots. Each GNF with an FPC set is seen from the outside and functions as a completely separate router, with traffic between GNF can be sent through the factory. This approach eliminates internal contradictions between the functions of the router, increases the stability of the system as a whole, and besides it has the “natural” ability to divide administrative management domains between organizational units. Allow one more illustration, they began to draw them very nicely.
The soon appearance of a universal chassis, where interface cards from PTX10008, PTX10016, QFX10008, QFX10016 and MX can be used, and this is what is reported in a press release from Juniper Networks, causes only one question - And what was it possible? Juniper already has some developments in the field of unification, for example, some of the MX240-MX960 line card hubs can be used on the top MX2010 and MX2020 for a long time. The current task looks more serious, since in one chassis it is necessary to combine the maps designed taking into account the peculiarities of work in all network segments - data center, core, boundary. How this is done is not reported yet, but according to Donyel Jones-Williams, the new platform is already in test operation in the networks of large customers.
So what's the future of trying to bring Juniper Networks closer? The manufacturer concentrates not only on the equipment, but mixes the accents in the direction of control and flexibility of operation. According to various estimates, the cost of the final product is only 40% determined by the hardware, about the same is the cost of software development, integration and automation tools. Let's see what happens and, if possible, we will take part in the development of the industry, because no one knows where the staircase will lead, as Doc Brown said - the future has not yet been written.
The press release and documentation on Junos Node Slicing can be viewed here:
→ Juniper Networks Unveils Cloud Grade Networking
→ Junos Node Slicing Feature Guide
Source: https://habr.com/ru/post/334684/
All Articles