Juniper PTX series as a step into a bright future
Juniper announced a new platform on the new Super-duper Express chipset. The PTX platform is called. There are two models: the PTX5000 shown in the picture and the PTX9000 that is not shown, which also scrolls the mouse wheels on the screen in two, but also twice as wide.
Do not worry, dear habrovchane, this is not a press release on behalf of Juniper Networks, urging to urgently buy a new product. They will cope with their advertising somehow themselves. Yes, and there is nothing to buy: announced is not for sale. Not to mention the price.
')
It makes no sense to list the performance indicators, count the number of such and such ports. Looking at the picture and the accompanying pathos, it is already clear that the numbers should look completely cosmic: interfaces start from 10GE, of which there are hundreds, several performance terabits per slot and all that jazz.
With functional more interesting. From the box, they threw out all the functions of aggregation of something. On the one hand, all routers turned out to be a router, on the other (in terms of functions) - in general, such an almost switch. They call it: converged supercore switch. Well, about converged, we’ll omit this for now, but let's talk about supercore and switch. The idea is that this is a pure P MPLS router. It has a minimum of simple IP functions, which, according to the idea, it needs only to provide connectivity for the control plane MPLS protocols. Well, PIM for nuclear multicast is still, because many people still disdain to poke it in MPLS. In general, no L2 / L3VPN and pseudowires, no VPLS, it is not even clear whether BGP. Well, probably still will. Queues, of course, only on the physical port.
That is, in general, more like a switch. With the only exception that, of course, there will be no Spanning Tree in it, by the night, of course, as well as any other L2-functions and protocols. But the Forwarding Plane scalability will surely be as cosmic as the performance. As well as support for all sorts of P2MP LSP, TE, Fast Reroute and other protections in full.
Surely on this new Express chipset, you can easily implement all operations with labels and without labels, necessary for the aggregation of anything. Moreover, it will be so when the same chipset is used in other platforms, you'll see.
But in PTX this is not done deliberately, including in order to save on the cost of development, support, and only partly iron. After all, the price of such units is not determined by the cost of the equipment. The point here is primarily in the complexity and number of functions, the requirements for the emergence of new features, the amount of testing, the number of potential breeding grounds for bugs, etc. In addition, the more complex the box, the more expensive its operation, the higher the demand for new complex features, the more users desire to shove it somewhere, where it is not necessary, to adapt the functionality to what it is not necessary to adapt, and so on.
The task was clearly to make a relatively stupid MPLS router for the kernel with as transcendental performance as possible (or is it now correct to say "cloud performance"?), Throwing out everything that can be thrown out of it.
It is known that a similar box is maturing (or has already been announced - who will tell?) In the depths of Cisco. The case clearly smells of a worldwide conspiracy. Trend, that is.
You want to laugh, or not, but in telecommunications a turn is coming on a new round of evolution. In IT (in the narrow sense of this concept), everyone rushes with clouds, drives computational power into huge data centers, computers change them to blade farms, and hard drives to SAN. In a word, it comes back to mainframes. All because mainframes now have something to connect with each other and users. Telecommunications are catching up. Do you think it's easy to drive your YouTube from America and back in such numbers?
The trend, of course, was predictable and started a long time ago. But really the first swallows fluttered up just now. Quite recently, manufacturers began to demonstrate the virtualization of switches to thousands of ports for megadatacenters. Juniper did it just a few days ago (more about that another time). Now here's the megahighway. An obvious hint at the world revolution, which, in fact, Pradeep Sindhu, chief technical officer, vice president and co-founder of Juniper Networks, is broadcasting from the corporate site what day it is. In its connotation (and a very loose translation) this is called saving the world from the exponent.
Everything returns to normal. Hierarchy in the small - in the trash, return the hierarchy in the large. Even the switching of the channels back was recalled: OTN interfaces were declared on PTX (yes, converged). It's only the beginning! Remember what machine switching is? “The locomotives were standing and pushing crowbars. Scrap with scrap connected - there was a switch. " Looking at the unit from the picture, it seems that the locomotives are no longer so far away.
Do not worry, dear habrovchane, this is not a press release on behalf of Juniper Networks, urging to urgently buy a new product. They will cope with their advertising somehow themselves. Yes, and there is nothing to buy: announced is not for sale. Not to mention the price.
')
What is it?
It makes no sense to list the performance indicators, count the number of such and such ports. Looking at the picture and the accompanying pathos, it is already clear that the numbers should look completely cosmic: interfaces start from 10GE, of which there are hundreds, several performance terabits per slot and all that jazz.
With functional more interesting. From the box, they threw out all the functions of aggregation of something. On the one hand, all routers turned out to be a router, on the other (in terms of functions) - in general, such an almost switch. They call it: converged supercore switch. Well, about converged, we’ll omit this for now, but let's talk about supercore and switch. The idea is that this is a pure P MPLS router. It has a minimum of simple IP functions, which, according to the idea, it needs only to provide connectivity for the control plane MPLS protocols. Well, PIM for nuclear multicast is still, because many people still disdain to poke it in MPLS. In general, no L2 / L3VPN and pseudowires, no VPLS, it is not even clear whether BGP. Well, probably still will. Queues, of course, only on the physical port.
That is, in general, more like a switch. With the only exception that, of course, there will be no Spanning Tree in it, by the night, of course, as well as any other L2-functions and protocols. But the Forwarding Plane scalability will surely be as cosmic as the performance. As well as support for all sorts of P2MP LSP, TE, Fast Reroute and other protections in full.
Why is this?
Surely on this new Express chipset, you can easily implement all operations with labels and without labels, necessary for the aggregation of anything. Moreover, it will be so when the same chipset is used in other platforms, you'll see.
But in PTX this is not done deliberately, including in order to save on the cost of development, support, and only partly iron. After all, the price of such units is not determined by the cost of the equipment. The point here is primarily in the complexity and number of functions, the requirements for the emergence of new features, the amount of testing, the number of potential breeding grounds for bugs, etc. In addition, the more complex the box, the more expensive its operation, the higher the demand for new complex features, the more users desire to shove it somewhere, where it is not necessary, to adapt the functionality to what it is not necessary to adapt, and so on.
The task was clearly to make a relatively stupid MPLS router for the kernel with as transcendental performance as possible (or is it now correct to say "cloud performance"?), Throwing out everything that can be thrown out of it.
Why all this?
It is known that a similar box is maturing (or has already been announced - who will tell?) In the depths of Cisco. The case clearly smells of a worldwide conspiracy. Trend, that is.
You want to laugh, or not, but in telecommunications a turn is coming on a new round of evolution. In IT (in the narrow sense of this concept), everyone rushes with clouds, drives computational power into huge data centers, computers change them to blade farms, and hard drives to SAN. In a word, it comes back to mainframes. All because mainframes now have something to connect with each other and users. Telecommunications are catching up. Do you think it's easy to drive your YouTube from America and back in such numbers?
The trend, of course, was predictable and started a long time ago. But really the first swallows fluttered up just now. Quite recently, manufacturers began to demonstrate the virtualization of switches to thousands of ports for megadatacenters. Juniper did it just a few days ago (more about that another time). Now here's the megahighway. An obvious hint at the world revolution, which, in fact, Pradeep Sindhu, chief technical officer, vice president and co-founder of Juniper Networks, is broadcasting from the corporate site what day it is. In its connotation (and a very loose translation) this is called saving the world from the exponent.
Everything returns to normal. Hierarchy in the small - in the trash, return the hierarchy in the large. Even the switching of the channels back was recalled: OTN interfaces were declared on PTX (yes, converged). It's only the beginning! Remember what machine switching is? “The locomotives were standing and pushing crowbars. Scrap with scrap connected - there was a switch. " Looking at the unit from the picture, it seems that the locomotives are no longer so far away.
Source: https://habr.com/ru/post/114887/
All Articles