×

Loading...

Object Oriented Network (1)

本文发表在 rolia.net 枫下论坛Every so many years, when some problem imposes a complexity so complicated for our poor little human brain to consume, another layer of abstraction and another layer of automation will come to the rescue.

That's happening in the network world today.

Back in the programming world, Object Oriented A/D/P (analysis, design, and programming) is not something new. Actually it was quite old if measured by IT time. It has becoming popular since the early 90’s, when I was a still a undergraduate in computer science major. The concept was to use another model to provide a different kind of abstraction when procedural based programming methods encountered a bottle neck due to ever-increasing complexity in the problem domain. Either it succeeded or failed, it was beyond this discussion. But defintely it provided an alternative approach.

In the network world, different layers of abstraction have been provided by many mechanisms, including the well known and well conceived OSI Seven Layer model, well accepted and well adapted TCP/IP protocols, and all the routing protocols, firewall rules, and application load balancers, WAN optimization boxes, etc etc. It’s been going well for more than 20 years, until now.

Now, the network world is facing some exceedingly complex challenges:

- Virtualization of the server world has revolutionized how computing is done. With the world’s 70+% work load is being performed from virtual machines, with the requirements to extend network visibility and optimization inside millions if not billions of hypervisors, with the notion of moving a virtual machine wherever a resource is available and doesn’t care layer two layer three boundaries, the complexity increased exponentially;
- When the demand to make network provision change within minutes to accommodate new application change, the Assessment=>Design=>Modeling=>Schedule a change window=>Make Configuration change=>SIT/UAT=>Post change support paradigm just doesn't work;
- The cost to maintain a complex network and make it work has ever been increasing since the Internet came into being. Many companies are feeling the pain and looking at ways to cut this cost (of course, including that many CCIE’s salary);
- The CLOUD, right, the cloud, the huge, highly condensed data centers and data centre interconnect. The well established layered hierarchical design model doesn’t apply any more. With a flattened network, with millions of host based routing, with the requirements of ultra low latency for east west traffic, with the need of no more than two hops away anywhere in the data centre, everything is gonna change or has been changing.

Apparently, another layer of abstraction is badly needed - with this new abstraction will sure come another round of automation. So came all the fuss of SDN, Overlay model, and Application Centric Infrastructure.

to be continued.,...更多精彩文章及讨论,请光临枫下论坛 rolia.net
Sign in and Reply Report

Replies, comments and Discussions:

  • 工作学习 / 科技领域杂谈 / Object Oriented Network (1)
    本文发表在 rolia.net 枫下论坛Every so many years, when some problem imposes a complexity so complicated for our poor little human brain to consume, another layer of abstraction and another layer of automation will come to the rescue.

    That's happening in the network world today.

    Back in the programming world, Object Oriented A/D/P (analysis, design, and programming) is not something new. Actually it was quite old if measured by IT time. It has becoming popular since the early 90’s, when I was a still a undergraduate in computer science major. The concept was to use another model to provide a different kind of abstraction when procedural based programming methods encountered a bottle neck due to ever-increasing complexity in the problem domain. Either it succeeded or failed, it was beyond this discussion. But defintely it provided an alternative approach.

    In the network world, different layers of abstraction have been provided by many mechanisms, including the well known and well conceived OSI Seven Layer model, well accepted and well adapted TCP/IP protocols, and all the routing protocols, firewall rules, and application load balancers, WAN optimization boxes, etc etc. It’s been going well for more than 20 years, until now.

    Now, the network world is facing some exceedingly complex challenges:

    - Virtualization of the server world has revolutionized how computing is done. With the world’s 70+% work load is being performed from virtual machines, with the requirements to extend network visibility and optimization inside millions if not billions of hypervisors, with the notion of moving a virtual machine wherever a resource is available and doesn’t care layer two layer three boundaries, the complexity increased exponentially;
    - When the demand to make network provision change within minutes to accommodate new application change, the Assessment=>Design=>Modeling=>Schedule a change window=>Make Configuration change=>SIT/UAT=>Post change support paradigm just doesn't work;
    - The cost to maintain a complex network and make it work has ever been increasing since the Internet came into being. Many companies are feeling the pain and looking at ways to cut this cost (of course, including that many CCIE’s salary);
    - The CLOUD, right, the cloud, the huge, highly condensed data centers and data centre interconnect. The well established layered hierarchical design model doesn’t apply any more. With a flattened network, with millions of host based routing, with the requirements of ultra low latency for east west traffic, with the need of no more than two hops away anywhere in the data centre, everything is gonna change or has been changing.

    Apparently, another layer of abstraction is badly needed - with this new abstraction will sure come another round of automation. So came all the fuss of SDN, Overlay model, and Application Centric Infrastructure.

    to be continued.,...更多精彩文章及讨论,请光临枫下论坛 rolia.net
    • 思科的DATA-CENTER SOLUTION好像已经解决大部分问题了啊,比如NEXUS 1000解决的虚拟机之间数据流的performance问题,两块SUPER板解决了升级当机问题(从65系列开始),VPC和FiberPath也基本上干掉了STP使得2层收敛更快。这篇东西哪里看到的啊?
      • 这篇东西是我自己正在琢磨的,见笑了。思科确实一直在做,有很多进步。两周前参加了Insieme的SE training。思科在这个问题上面临的挑战是多方面的,包括一些传统“软件”公司。My take is "piece meal won't work...there will a total design paradigm shift..."
        因为同为一类技术, Fabric Path 也许会被VxLAN 取代。
        • 说的是,我好像哪里看到过基于应用流交换的下一代网络,找找看
          • 找到了,software defined networking. 基本上这个东西成熟了,所有的传统路由交换IE和NP们可以下岗了
            • 无论名字叫什么,SDN, Overlay, 还是APIC, 最终各个vendor都意识到这是个机会,软件公司可以进入一个全新领域,Server公司能扩展其private cloud offering, Hypervisor公司能变成网络公司,对前3者,这场变革只有好处,没有坏处。而对于思科,则是个考验。
              • 按照钱伯斯的思路,他会先观望走在前面的几个公司拼杀,等拼的相关技术比较成熟了,挑一家性价比高的小公司收购,把产品重新包装通过强大的渠道用低价打市场,把同领域的竞争者们干的差不多了然后提价卖产品和服务,等到市场饱和的时候裁员降成本。希望这次他不能得逞。