Wednesday, December 04, 2013

Hypervisor based Overlays or Web Server based Overlays?

There  is a characterization of overlays as an abstraction as defined in computer science. An abstraction can be compiled down to the physical sometimes using an intermediate format. (like objects to byte code). Overlays are not abstraction, they are mark-ups and create a new layer for processing.  Mark-up are hints to higher layer software on how to process the content. There are services that these overlays need from the underlay and the exchange points are implemented in gateways.

Lot of complexity is being built into overlay networking because the design center of overlay networking is the hypervisor. By making hypervisor as a key node in the federated controller architecture, we are leaving out alternate mechanisms that enforce tenant/workload isolation like lpars/containers etc.  which don't need or require hypervisors.


Wednesday, September 11, 2013

Packets, Flows and Messages

I came across this post while doing the daily reading on SDN  https://devcentral.f5.com/articles/the-bifurcation-of-the-network-flows-versus-messages#.UjDVZ7Hn_mE. This resulted in a flashback as I recall sometime in 2004/5 time frame I was trying to explain to folks that message based switches are the way to go for then emerging service oriented network.

I would articulate the messages vs. tuple (flows) vs. header (packet) a little differently. In a service oriented network (and it is not SDN), the brain or intelligence bubbles up into a overlay network of proxies. Those proxies communicate among each other using messages. The brawn sinks down into a dumb data plane. The mapping between the two is done by "binding" to a transport. In those days HTTP was the transport but it was very apparent that it could not serve the purpose due to lack of asynchronous MEP (exchange). Today we are closer to realizing that vision.

Alas.. but we got hit on the head with murphy's law. In came SDN trying to disrupt the data plane by creating a centralized control plane. Except the control plane is not as intelligent as a the control plane of the original SON concept. It still works on packets. Fortunately, the byproduct of SDN hype is the real fruit of all this labor. That by product is programmability of the network device (not the network just the device). Today a message processor can read a route sent over the overlay of proxies and program the control plane on the switch to change the flow direction. This was not available in 2004/5. This was the original network virtualization concept. Virtualization because in the message processor you can have multiple tenants (although in those days we called it applications) and each one sees a different topology for its packets.

Big learning from that experience ... it hurts more to be early more than to be late.

Monday, September 09, 2013

Using Options Exchange to Price Cloud Computing

Traditional business decision tools like ROI, NPV do not sufficiently capture the risk in an investment decision on cloud computing. The demand for cloud services is not a sales forecast (linear) but more like a stock price (stochastic). Today this demand is met with a service that prices itself using an ancient cost accounting method, resulting in pricing models like pay-as-you-go or all-you-can-eat. The price does not take any market input. Imagine buying the stock of a company for a constant price regardless of the volume.

An option is an instrument that give you the right to buy the underlying security but does not obligate a purchase. Trading this instrument with underlying compute capacity as the security will help researchers and businesses study the demand for the cloud computing. Knowing the real demand and the market price for this emerging commodity will help guide investments for service providers and businesses. In the absence of this, we are going down the monopolistic pricing path that we are seeing right now. Except in the case of cloud computing, there is no government backstop for this monopoly.


Monday, August 12, 2013

SDN: Technology and Business Model

If SDN is an architecture, then what is the technology? Clearly it is not virtualization, in fact, most SDN architecture are not yet virtualized. It is not a protocol. The whole point of giving software the control is not to be constrained by a protocol. It is the API. API is what makes SDN possible. APIs like a network stack is not monolithic. There is the free API i.e .REST and there is the premium API which may or may not be proprietary - but will always be Open. APIs come with their own infrastructure and scaling that infrastructure is the challenge of SDN.

Moving on to business model of SDN. It is not in selling controllers. There are too many of them. It is not in scripting, integrating and scaling them in a cloud infrastructure like say OpenStack. Those are what is normally called out as NRE in the ROI spreadsheet. The business model of SDN is in hosting the cloud. SDN lowers the barrier to entry to enter the business of hosting.

So if you want to make money off SDN, get into hosting. If you want to work on SDN, get into APIs.

Sunday, May 12, 2013

Internet of Things

There isn't just one web. There are many webs, the one we call web is the information web that can be browsed. There is a web being built that is invisible to a browser, it is the web of machines - referred to sometimes as IoT. As usual it is being met with cynicism, because we are not able to see the potential of machine to machine communication. Machines don't have to be physical devices, they can software logic elements. The communication transport does not need to be heavy weight, it does not even need to run entirely over existing TCI/IP transport. Some of the possibilities of IoT are obvious i.e. call home feature, automated trading floors, intelligent surveillance etc. But some are not so obvious. Take for example, your TV prompting you to watch a program because it read a note on your blog or read your profile and "understood" that some program running now is important to you. (apple tv take hint please). Or it knows using face recognition using kinect who is watching the Tv and adjusts the shows according to preference.

I would like the kitchen not just the refrigerator to tell me that the dish on today's cooking list does need an ingredient that I don't have or will run out of soon.

How about my shingles on my roof detecting the leak and sending their location so I don't have to spend hours on the roof with a bucket of water to find it.

Or how about my favorite.. ,(being the only guy I know who had to lost two cars to engine blow out because the engine did not enough oil and broke down)  where my car simply decides that maintenance can no longer be postponed and using  the yet to be developed or perfected driverless technology drives itself to a oil changer get its maintenance after dropping me at work.

Increasingly the innovation is in removing the human element from processes which ultimately makes the human more productive.

Saturday, March 30, 2013

Cloud Requires a Market Maker

The business model of cloud computing is boiling down to keeping marginal cost of providing capacity to a marginal demand from mostly corporate user. Even though the cloud enthusiasts continue to believe all corporate computing will move to the cloud, it is not happening and unlikely to happen. The marginal demand though is moving to the cloud and many corporations are designing an overflow capacity in the cloud (kind of like line of computing account). Serving this marginal demand with capacity at a marginal cost that covers the service provider margin requirements is becoming the biz model of the cloud. We have seen the ramification of this on data center site selection and facility design. Now the ramifcations are becoming apparent in the infrastructure that gets deployed in these data centers. Few standouts are ...

Fungible is perhpas the most important requirement for any infrastructure that will be deployed in a cloud. No bells and whistles is another requirement. Basic bare bones is winning over hardened and over engineered systems. The datacenter managers use software to fill the gap between the skeleton and skin. Another standout is virtualization is not a  key requirement. In fact, as more applications onboard it is becoming apparent that cloud based service has to be hybrid of virtual and physical. All of these put together still not enable datacenters to serve the marginal demand at the lowest cost in almost real time. I don't think this can be solved technically, it  is really not a technical initiative, it is a market initiate. Makes sense as demand and supply are really market forces.  Customers engaged in price discovery need a market to query and a market maker to take the order. You can look at airline industry or the securities industry to get a feel for how this might evolve. Consolidators buy blocks of tickets at steep discount in anticipation of demand. Market makers take position in stock with similar intentions. In both of these industries the market maker is a critical element without whom the industry would not function.

This is what we need in cloud industry as well. Something or entity that plays the role of a Goldman or Priceline. I am waiting for the day when AWS starts a program which uses the rack of computer in my garage as capacity to sell to their customers. Kind of like the smart grid or SETI program. We may after all realize the vision of grid computing. The missing link is the market maker.  

Thursday, March 28, 2013

Industrial Internet

May be they should rename industrial internet as software defined manufacturing or at least smart manufacturing. There is lot of software in manufacturing already mainly driven by precision requirements. The name, Industrial Internet, conjures up notions of a network hardened against hackers and natural calamities. Moving to software defined everything SD-* is essentially accepting that most basic intelligence today can be codified and stored in data bases that can be queried at high speeds. But that does not mean there is no room for software learning.

Software learning is where software does not just query for data but also reasons with the results of a query and draws inferences and acts on them. This type of software should form the base platform of industrial internet or just plain old IoT. When the devices on you connect to one another not just at the bluetooth layer but at a application layer. For example, when a device playing music is in the vicinity of a speaker, it switches to the speaker (after asking of course) or when you suddenly brake your car to avoid an accident, the video and music playing devices inside the car stop to let the passengers focus on the road, or the same behavior when listening or watching movies in an airplane.


Costs in Training LLMs

 I went through the Llama-2 white paper that was released with the model by meta. I was hoping to learn some special technique they may be ...