The last few months have been a blistering pace at Plexxi and it has impacted my time to write. Writing is important to me as it is my method of thinking in depth without the interruptions of email, calls, text and tweets. Outside my window a Biblical rain is falling and I have Zac Brown playing. As with past notebook entries, here is collection of topics I have been reading and thinking about over the past few weeks.
I read a half dozen analyst reports concerning Cisco’s Analyst Day. Most of the reports bullish on CSCO’s prospects for 2013. The headlines regarding CSCO M&A plans were interesting on the Friday of the meeting. At first the headlines said “Cisco says company has gone too long without big M&A, expects to be active in it.” I read that and thought CSCO needs to buy HPQ. They could sell the server business to DELL, sell the PC business to Lenovo, put a red shirt on all the Autonomy employees, sell the printer business to Lexmark and close networking, keep software and EDS. Later in the day the comments around big M&A were clarified, toned down and apparently the HPQ deal is not going to happen…:-(
If it was January 2003 and someone asked you which of these four stocks would perform better over the next ten years, which company would you choose from this list: EMC, CSCO, DELL and HPQ? Here is a link to ten year chart. To save you from clicking out the answer at the time of writing (which was 12.10.12) is: EMC +274%, CSCO +32%, DELL -62%, HPQ -26% and the SPX is +51%. I think those numbers are incredible. CSCO has underperformed the SPX benchmark by 1,900 bps. Coming off the collapse of the 1990s bubble, EMC made a focused effort to be a software company. People come to work everyday and create stuff. That stuff needs to be stored, replicated, organized and backed up. CSCO has made a lot acquisitions in their history; looks like 152 for ~$66b. They seem to have done better with the larger deals (e.g. Stratacom, Starent, SA, Webex, Cerent, IBM Networking Div and Andiamo), but there is no question that Crescendo was the first and most important deal in the company’s history. With the reports over the weekend that Cisco was looking to sell the Linksys business, it does make one wonder if the era of vertically integrated technology companies is once again beginning to wane. Does CSCO need to get into more of the software business?
SDN – Start of CSCO vs EMC?
A colleague mentioned the idea that SDN seems to be the start of EMC versus CSCO. When VMW purchased Nicira, in my mind that was play against the installed base of networking gear. I read one analyst report this morning that said CSCO had a $181B installed base. The sales pitch around Nicira is what I call a brown field sale — it is not the next network generation beyond switching, it is something you sell to people who have a lot of VMs and poor switched network performance. Nicira shines in the brown field with an overlay network. The pitch is an additional pay as grow spend that overtime saves on the combined costs of CAPEX+OPEX. It has nothing to do with cheap white box switches.
I wonder what the Insieme team is up because they have to be in development of a controller. Earlier in the year I envisioned it a mediation controller between NX-OS and the rest of the world. It probably allows better orchestration of the Insieme domain which I think includes storage in a leaf/spine architecture. If SDN is the starting point of EMC versus CSCO war, does that mean CTXS is probable acquisition target for Cisco? CTXS is probably too big for CSCO to swallow, but if they did bid with the plan to spin off the Online Services (GoTo business), it is possible to have a $2-2.25B competitor to VMware.
At Plexxi we put a lot of content up on the web page about two weeks ago. If you have been a regular reader of my blog, I would assume that the older content now falls into place. When I wrote this post in June 2012, it was veiled description of Plexxi’s first generation product set. We are well into our second generation product set. As we near end I would like to thank the many customers we have engaged with over the past year that have helped us define and accelerate our second generation products, but I will leave that for 2013. There were five important bullets from my June 2012 post on SDN. I can now translate what these bullets mean in relation to Plexxi.
- Computation and algorithms, in a word math. That is what SDN is.
- Derive network topology and orchestration from the application and/or the tenant. This is how SDN is different from switched networks.
- Ensure application/tenant performance and reliability goals are accomplished. This is part of the value proposition.
- Make Network Orchestration concurrent with application deployment. This is what SDN is for.
- A properly designed controller architecture has the scope, the perspective and the resources to efficiently compute and fit application instantences and tenants onto the network. This is how SDN will work.
1. “Computation, [commodity flow algorithms, graph theory]; in a word math. That is what SDN is.” That is what Plexxi does with our controller called Plexxi Control. We compute topologies using relationships called Affinities. You can find a white paper on Affinity here.
2. “Derive network topology and orchestration from the application and/or the tenant. This is how SDN is different from switched networks.” We do not use distributed protocols to figure out state. We calculate state on each switch, which is why we have a high-performance co-controller on the switch. This leads to something that we call multi-core networking. Here is a white paper and here is a link to something I wrote back last summer on the dawn of the era of multi-core networking.
3. “Ensure application/tenant performance and reliability goals are accomplished.” This is how Plexxi uses an API and controller. In a Plexxi network, the network can be orchestrated. It can be diurnal. When I talk to network practitioners, I tell them how they can have a network configured one way during the day for knowledge workers and another way at night for replication, backup, compliance, computation, etc.
4. “Make network orchestration concurrent with application deployment. This is what SDN is for.” With a Plexxi network, we have a dynamic and orchestratable physical interconnect. We can direct flows using optics, bypassing switching silicon as needed. When we configure a Plexxi network, this can be done top down from Plexxi Control. We fit the network together based on the needs of the application. This a very different way of looking at the network. I have been in networking for a long time and when I talk with network practitioners, we all design networks through the same approach. We first start drawing connectivity between the things we want to network. In a Plexxi network, we do that too in a default configuration, but we use Plexxi Control to define and compute affinities that allow the network to be reconfigured to meet the needs of the applications. An application architect can tell a Plexxi network what it needs. When I use the word application, I want to be clear that to point out I am not talking about network services. I think load balancers, firewalls, routers….those are services that exist in the network. They are not applications. Applications are software and in a Plexxi network software can tell the network how it would like the network configured. Plexxi Control then computes whether these demands can be fulfilled.
5. “A properly designed controller architecture has the scope, the perspective and the resources to efficiently compute and fit application instantences and tenants onto the network. This is how SDN will work.” This is a very broad discussion topic on state, controller architecture and deployment. In a Plexxi network we have a central, out of band controller (i.e. Plexxi Control) for topology calculation. We also have a high-performance co-controller on the switch for state calculations, but it is not dependent on existence of Plexxi Control. Plexxi Control provides the scope to the network, while the switching and state calculations are done at the individual switch level.
In the past week Martin Casado of Nicira spoke about their transition away from OpenFlow. To save you from clicking out, here is the ket quote “OpenFlow says the world should like this: You have this table that has an 11-tuple look-up, which is this super-general thing, and you have a whole bunch of them,” Casado said. “In order to get the OpenFlow checkbox, a lot of vendors will simply overload one of these tables, which will have maybe 5,000 entries. And they try to shoehorn OpenFlow there. These chips were actually not made to do that. OpenFlow is still trying to adjust to this, but it’s going to be a very difficult thing.”
On a Linkedin comment tread to the above referenced article, the CEO of Contrail commented:
- Ankur SinglaInstead of all the noise around the acquisition, I would like to state the following: Contrail’s Network Orchestration Systems provides Network Virtualization and Networking Services without requiring Openflow in either Virtual or Physical layer. In addition, we believe that Openflow in Data Center Switching Fabric/Hardware makes no sense because of many reasons – we will blog on it!
- Ankur SinglaThe compiler approach that Kireeti blogged about has nothing to do with overlays – it is about the abstraction that is presented to the operations and app developer. Contrail solution does tunneling/overlays from the hypervisor or bare-metal kernel driver.
I have had some horrible experiences with the Cloud over the past year. Gmail was down this morning (written on 12.10.12) and other various social sites a few times over the past year. The worst experience I had was with iCloud, which was a total failure. I needed to restore my iPhone from my iCloud backups and it turns out my last four or five backups were corrupt. I had no way of knowing that they were corrupt until I needed them. That is clearly an iFailure, because if I was backing up locally I could not backup to the iCloud. I tried to use Plaxo to sync contacts across multiple platforms, well that did not work. I give them props for trying, someday some group will make contact syncing work across all devices, computers and clouds…those will be really wealthy people. I am now back to using Gmail contacts, synching to my iPhone with the $3.99 app. It works really well and I use the cloud on all computers. IMO the cloud still fails too much and it is still DIY. When it is working it is great, but when it fails it is 10X worse than it is great.
Long: GOOG, CTXS, LNKD, AMZN, GLD, AMT and CCI.
The Web We Lost
I found this post on the web we lost interesting. I think the comments are even more interesting to read, but IMO I really dislike ads and business models designed to deliver advertising, which I have written about in the past. When I am using a site or an app that has ads, I want to use the ad free version or I just stop using it. If the web becomes all about delivering ads to my user experience, my participation rate will decline. See today’s news on Instagram.