Hypothesizing About the Future of Networking
I am off to CA for the week, so posting will be infrequent, which of late is par for the course. With earnings beginning, I would like to offer some thoughts, but most likely it will have to wait till the weekend. One daily activity that I was consistent about early in my career was keeping a notebook. I have a storage bin in my basement full of notebooks. In these notebooks are network designs, customer meeting notes and internal meeting notes across five companies. I became inconsistent and lost the skill of keeping a notebook in my thirties, but a few years ago I read George Soros’s Alchemy of Finance.
The book made me remember the value of keeping a notebook; a short list of my day’s activity with reminders to follow-up, random ideas to research and a lesson learned or valuable observation to remember. For the past few months I have been writing this blog somewhat from my daily notebook. I have been asked many times what SIWDT means. It was an acronym from my notebook for stuff I would do today. When I started the blog, I had every intention for it to be online, public view of stuff I was doing and I did not care if I made mistakes, got stuff wrong – that is life and all part of learning. I used to keep a hard copy notebook and for me there is a personal preference to the tactile aspect of a notebook and pen – but I have adapted to using an on-line document that allows me to paste links, emails and web content.
I have been spending most of time looking at and engaging with people who operate service provider networks, enterprise networks and data center networks. I do not know if my predictions will be correct, but that is why I keep a notebook and that is why I have a blog which is a distillation of notebook accessible by all. My notes help me to remember, to check, to confirm, and to look for supporting evidence of hypothesis A or B and it helps me frame the relevancy of a data point.
When I look at the three groups of networks (i) service provider, (ii) enterprise and (iii) datacenter, I see the future of the network structured differently than how it is today and my interactions with the vast majority of technology people lead me to believe they are in a state of denial when comes to pending changes in network design and what will be important in the future. This has been a reoccurring theme in my blog over the past few months. I have high level of confidence in my hypothesis because the trends and data points continue to confirm and support. I go back to my notebook and check on reoccurring data points and when I participate in a series of meetings with disparate entities and I see and hear the same objectives stated differently, I feel confident of direction. In the past few weeks I have seen network people at a MSO, PTT, a financial services company; a data center operator and an ILEC all present a desire to reclaim bandwidth in the network. Primarily they all want to use bandwidth they have allocated for network protection to be used actively. In four of these scenarios the traffic amount was very large. I remember one meeting at service provider in which a person diagramed how they wanted traffic flows to occur across their nation network; describing the desire for some flows to stay local and others to move across the backbone. To me this was a familiar request because I had seen the same from datacenter person wanting some flows to stay within racks or clusters and not transit backbones. Same architecture and network flow request – just different geographic challenges.
When I look at the technology direction in the datacenter (DC), I see it differently than most incumbent networking companies. I see a desire to flatten the network; to remove the vertical stacking and leaf/spine network architectures. VMs need to be software configurable and portable on the network element. In many aspects, I see the architecture and technical demands emerging from the modern day DC to be the new leaders that will influence how networks will be designed in the future. I view the end of the client/server era to be May 2001. It was in May 2001 that RLX Technologies shipped the first blade server called the RLX 324. That was the event that started the transition from the client/server era to our contemporary datacenter era in which the large commercial datacenter is possible. With the rise of concentrated compute, increased storage and virtualization, we have a new technology era – except the network failed to evolve. We are very much building networks that look like networks from 1995-1999 – the client/server era. I know this because I can see them in my notebooks from the 1980s and 1990s.
With leadership flowing from the DC, that is why we see the rise of applications like Hadoop which are designed to split up work groups and distribute data intensive processes across many servers. I look at this application and think it is only the beginning and five years from now a variation will be running in service provider networks serving the needs subscribers for content (e.g. video) and games. It is not stretch to equate the original Akamai CDN service to a contemporary Hadoop cluster. The scale and application was different, but the distribution and division of content was not too far apart in terms of the end result. As the modern era of the DC emerges, we are now seeing a convergence around common IT metrics and I dare say we will see a flatter network in the DC and that same evolution will begin in the service provider network in ~5 years. The DC network leads the enterprise network and ergo it will lead the service provider network.
The network drivers have been and continue to change. Complexity will be pushed out of the network and into the I/O and the CPU. Moving network complexity to the CPU element is the future because it is the CPU element that is most critical to content and applications. The I/O is the next point of statistical gain. The network will become simpler and tools of the network operator become the tools that allow applications to be managed, measured and servers and VMs clustered. Scaling software to manage these elements as network capacities grow significantly over the next 2-10 years is where a lot of value will be created. In all, just some thoughts about the future and it is possible that I am incorrect.
* It is all about the network stupid, because it is all about compute. *
** Comments are always welcome in the comments section or in private. Just hover over the Gravatar image and click for email. **