An ecosystem for intelligent systems consists of several players, such as chip, OS, middleware, end device, cloud service, back office processing and analytics providers, and system integrators. Ayla Networks, which is still in stealth mode, claims that they provide secure connectivity for an end-to-end solution for an intelligent system. They currently focus on the consumer market but do not rule out expansion into other areas.

I sat down with David Friedman, CEO of Ayla Networks, during the recent Design West to find out what they are up to.

David Friedman

Who they are

David was VP business development for a wireless chip company before. After selling it in 2010 , he saw a business opportunity. At that time, end devices were beginning to be connected to form the Internet of Things. But the ugly reality was that those thousands of end devices were very different from each other, with microcontrollers in a variety of architectures and operating systems, as compared with the nonembedded world dominated by Windows and Linux. All those differences sure were a hindrance to accelerating and proliferating the Internet of Things. David and his cofounders saw the need for a generic solution that could absorb these differences. That led to the formation of Ayla Networks.

David and his team started to work on his solutions. Using his background as a chip guy, he teamed up with STMicro because ST is a major player in the microcontroller market. Ayla Networks is a software company and does not deal with hardware, so this is a good combination. Chip vendors focus on how to design and develop new and better chips but are not experts in networking such as Berkeley sockets and SSL. In other words, companies should focus on their core competency and outsource the rest. In the same vein, application vendors are not experts in the lower layers of software that support applications. During Design West, I heard from several players that application vendors should outsource the lower layers and concentrate on their core business; that is, design and develop applications.

So David is saying “Come to us. We will absorb any protocols differences and security needs to support your applications. You do not need to worry about the lower layers and other infrastructure concerns.”

What they do

Ayla provides an end-to-end connectivity software; for example, to remotely control your AC from outside your home with your smartphone. If you implement something like that on your own, you need to develop lower-layer software for the smartphone, including secure interfaces with its OS and networking stack. Then you need to develop an application to work with that infrastructure. Then you need to worry about how to connect it to your target AC. Communication can be via cellular, WAN, LAN, or PAN. You need to choose the right one. Finally, on the target AC, some mechanism needs to be incorporated to receive data and control from your smartphone. For that, a small board with a communications chip on it must be inserted along with the lower-layer software. And as with your smartphone, you need to interface with that chip’s OS and networking stack, on top of developing applications.

What Ayla provides:

1. Client-side lower-layer software for applications
2. Networking solutions with security
3. Cloud services
4. Lower-layer software on target appliances

The client-side software can be integrated with your applications and downloaded from Apple Store and Google Play like other applications. Ayla provides whatever networking protocols are required by the applications. In addition, they provide cloud services to connect your client devices to target appliances. David did not elaborate on how they provide such services. Cloud services consist of cloud infrastructures and applications in the form of virtual machines. Because of the proliferation of inexpensive cloud infrastructures services, a startup like Ayla can afford to provide the cloud services. The lower-layer software on target appliances is the same as #1. Application developers can focus on their core business of developing applications without getting bogged down in lower-layer stuff.

Team

Now this seems to require a lot of technical expertise in several areas, such as embedded systems, networking, and cloud. Although these areas are closely related, no one person could address all of them. Although David did not reveal details about his team, he did say that he gathered technical people who had worked together for several years. People who like to innovate and have a passion to create something new are attracted to his team.

Devil is in the details

Many people have discussed controlling an AC from outside with a smartphone or a tablet, and that by itself is nothing new. David told me that now is the perfect time to bring their solutions to the market. Technologies have advanced and the market is opening up. An article by Reuters reports that by 2022 a typical household will own 50 Internet-connected devices, compared with 10 now. David said that we do not want 50 solutions for 50 devices but only a single solution so that any new device can easily belong to the existing network. He also emphasized that creating a supportable product is really, really difficult.

Their infrastructure pieces must be easy to:

• install
• configure with a lot of latitude
• scale
• implement with secure delivery

They claim that they have met all four requirements.

Big Data

They are in a perfect position to collect and aggregate data, but David did not reveal any future plan for business exploiting such a position. But he did not rule out the possibility, either. If I were an AC OEM, I would be very interested in analyzing control data sent by smartphones, to reflect on how to tune my AC features. David told me that the key to the use of Big Data is anonymization with the ability to opt in or out.

Energy consumption

What about power consumption? Smartphones eat a lot of power, and additional features like these would consume even more. David told me that his developers pay a lot of attention to curbing power consumption. Power-use optimization implemented as in the iPhone would attain energy efficiency.

We chatted a bit about power in general when everything is connected. My view was as follows:

1. Advantages: There are many advantages to deriving useful information from generated data that might be otherwise discarded. Some information can be used to save power.
2. Disadvantages: Unless we can intelligently select which data to collect, or keep, or discard, we will end up with a pile of useless data occupying a lot of storage and server equipment, wasting energy.

I think what David said about the disadvantages was interesting. He said that analyzing a vast amount of data, transforming them to a small number of useful data, and discarding the rest might do the trick. I do not know the doability of such a thing, but it is an interesting thought.

Future

David did not give me any concrete future plans, but this system can expand beyond the consumer segment to the commercial and industrial markets. I think there is a reasonable level of traction

in the consumer market at this point, and there will be greater demand later. In addition to a clear application for turning an AC on and off, I can think of a few more examples. Sprinklers for lawns are usually on timers, and occasionally they start to work even in the rain while you are not at home. Your remote device can override this. Or better yet, you can program sprinklers in conjunction with moisture-detecting sensors buried in the ground and with sensors for other local weather.

But I think the really big applications are in the commercial and industrial segments. I think it is very smart of Ayla to choose the consumer market first. There are two reasons. The first is that the commercial and industrial segments are known to be late adopters. The second is that if you target very specialized and sophisticated industry-grade equipment, how many people will know? But familiar appliances like ACs show up on many people’s radar screens; after success in the consumer market, Ayla can enter bigger markets.

The conversation stayed at a high level because they are still in a stealth mode, but a public announcement is forthcoming. Meanwhile, you can register to purchase their design kit by visiting here.

What does it mean?

How does the IT scene change with the implementation of hybrid cloud computing? First let’s consider private clouds only. In the following, I will use an enterprise data center and its private cloud interchangeably for the ease of discussion, although not all data centers have been converted to private clouds yet. Some company may have several data centers (and therefore private clouds) in the US, or even worldwide, across multiple time zones. So even before talking about hybrid, using this technology we can combine those physical data centers into one single logical private cloud. A logical cloud consists of physical private clouds (data centers) and may be recognized as one entity.

Logical private cloud

With a logical private cloud, using some technologies from CloudVelocity, we can move applications that may consist of physical machines (PMs: not virtualized) and virtual machines (VMs) anywhere and anytime we choose. In the following figure, we can pass PMs and VMs back and forth seamlessly between our home cloud and any other private clouds of our company. Although it shows only a subset of interactions below, we can potentially move PMs and VMs in any way that makes sense by some predetermined criteria. It may that one PM or VM is passed to another cloud and then to the third one and so on. It would become pretty complex to manage your Pms and VMs under such a new paradigm.

PMs and VMs move around only among private clouds owned by the same organization. A set of such private clouds may be considered as one logical private cloud.

This means we can finally implement several things discussed in part 1, including:

• Follow the sun
• Follow the moon
• Load-balancing
• Fail-over/disaster recovery

Follow the sun

In a given workday, access to software applications and utilities running on servers and other IT equipment—and therefore clouds—fluctuates. Access starts to grow as people start their day’s activities in the morning, it hits a peak, then subsides towards the evening. Access is lowest during the night. So you might want to move your PMs and VMs to other time zones where the sun still shines and more loads need to be processed. We can expect a better response time when loads and processing units are close to each other.

Follow the moon

In many countries, power is cheaper during off-hours (normally nights, hence follow the moon). Sending your Pms and VMs to such time zones may reduce your operation cost. Additionally, even within the US, power cost can fluctuate hourly if a variable power pricing model is implemented and applied to data centers. By shifting your VMs to a data center whose region gets the lowest power cost, you may save on running costs.

Load-balancing

Just as we load-balance among servers at a data center, we may want to send loads to several different private clouds. In this way, when one data center gets very busy, such loads can be passed to other data centers to share the burden. How you move PMs and VMs should be determined by predefined metrics to optimize your operations for a few factors, such as operating cost, response time, and throughputs. Each organization has its own goal for its operation, and the metrics should be tailored to accommodate it.

Cloud bursting may be related to load-balancing, although it is not the same. When a load increases in a private cloud, we may want to move all or part of it to a public cloud for on-demand processing; this is known as cloud bursting. PMs and VMs that are processing the load can be moved to a public cloud for continuous processing. When the load subsides, PMs and VMs on the public cloud can be disabled. There has been a lot of talk about cloud bursting, but now it can become a reality. We need a good automated system to move PMs and VMs, and to enable and disable them as needed. A good policy is a must-have for this as well.

Fail-over/disaster recovery

The San Francisco Bay Area will have a major earthquake some day for sure, and when it happens, much of the existing infrastructure, including data centers, will be unusable. If we have a way of duplicating what we are running in our primary data centers at a secondary site far enough away (such as the Sacramento area, a little more than 80 miles from the Bay Area) and transferring execution state information intact to the distant site, processing could proceed without interruption.

Super logical private cloud

With this technology, we do not have to consider the boundary between private and public clouds either. So the logical private cloud can include public clouds, becoming a super logical private cloud, or what I call a supercloud.

A green oval depicts a private cloud, and a light-blue one represents a public cloud.

This configuration would make managing PMs, VMs and clouds much more complex. We can move our PMs and VMs between private clouds, between private and public clouds, and among public clouds. We will no longer be restricted to a move between one cloud and another cloud (a one-to-one move) but can implement one-to-many and many-to-many as well. Then it will become necessary to develop a system that allows automation. As we involve many private and public clouds of various implementations, we will not be able to easily track how to optimize such moves. For that, we will probably need a policy based on predefined metrics. Cost may be the number one factor. But at the same time, we want to maximize response time—and the performance of developers scattered around the globe.

Also, note that many superclouds may share the same private and public clouds. This means that loads at each private and public cloud could fluctuate over time. So depending upon how busy each cloud is, we may want to dynamically alter how we form a super logical private cloud for optimization.

By the way, when a supercloud is developed and deployed, will we call it a supercloud or simply a cloud? Those IT folks who will follow us in the future may take it for granted and consider it a normal IT deployment and execution environment. Throughout IT history, when some technology or method becomes transparent as part of an overall system, that is when we say that that technology really has matured.

Who uses hybrid clouds and benefits from them?

I can think of three parties, although there may be more.

Enterprise end-users

Enterprises that have their own private clouds can extend them to public clouds to produce hybrid clouds to exploit the things I mentioned above.

Data center providers

If you are a colo provider, you can sell extra services at your center to realize hybrid clouds for your clients. There are different levels of providers. Some may simply rent a space, while others provide both equipment and services. Some may provide both private and public clouds at the same data center. For them, this is a perfect tool to increase their revenue.

Third-party consultants/service companies

If a colo provider does not want to provide any service other than space, those guys with the hybrid cloud technology can help end-users implement hybrid clouds.

Energy consideration

Finally, my blog always ends with a question about what the subject means to energy efficiency. Although inconclusive, there has been some discussion about whether cloud computing is more energy efficient than its predecessors. I think it depends upon whose view you take. If you are a user, you pass some or all of your computing needs, along with support staff, software, hardware, power, cooling, water, and other things, to your cloud provider on an on-demand basis. Since you can reduce your investment on these, it is certainly energy efficient for you. It may or may not be for your provider. If the provider has very little utilization of their facilities, they may not be profitable or energy efficient at all. You may still have to have a large staff, a large space, dedicated IT and facilities equipment, facilities support such as cooling, and so on. That cannot be very energy efficient.

When a hybrid cloud becomes a supercloud and our energy becomes more scarce, we may need to look at energy consumption and energy efficiency at the supercloud level without distinguishing private or public clouds, which may sound silly at this point. It is because the US seems to be doing fine for the foreseeable future with shale gas and oil, but who knows what may happen next?

Some technical details

As a former technologist, I wanted to open the hood and find out more about the underlying technologies. For this, Anand Iyengar, CloudVelocity’s founder and CTO, gave me a chalk talk.

Anand Iyengar

Because this is not a white paper detailing the technology, I only describe it at my layman’s level. However, it is such an intriguing technology that I’m accepting Anand’s offer for further discussion and will write more about it in the future.

Anand elaborated on the details, but I made a simpler diagram to fit the space. It is not that much different from the picture above.

Virtual machines (VMs) move between a typical enterprise private cloud (mostly VMware-based) and a public cloud (typically Amazon AWS). (Source: CloudVelocity)

Setup:

Let’s take a quick look at the architecture:

Private cloud

• We first look at your own data center or colocation facility (private cloud). In the modern software application system, an application does not run on a single server. Instead, the running of an application spans multiple physical and virtual machines. So we call it a multisystem application. The configuration may differ according to usage and design. Typically, it consists of load balancers, web servers, application servers, and sometimes a cluster of other servers.
• This is illustrated in the figure above. To save space, I drew only two machines, S1 and S2. The multisystem application typically uses a database, file systems mounted from a closed-box NFS server system (NFS1), and services from an LDAP server (LDAP). Everything in the public cloud is a copy of what is in the private cloud, including NFS1. Note that NFS provides files locally but not over the cloud boundary. Moreover, in the private cloud there is a server, such as an LDAP, that one may not want copied to the public cloud but kept in the private cloud for security reasons.
• There are virtual appliances (CloudVelocity Nexus Site Manager for the private cloud and CloudVelocity Cloud Manager for the public cloud) that together keep the cloud site images synchronized with the most recent changes to systems in the private cloud. CloudVelocity uses the term appliance to emphasize its dedicated function. CloudVelocity Nexus may run on a physical server, while CloudVelocity Cloud Manager runs as a virtual machine.
• Let’s further assume that S1 (in the VMDK file format) is virtualized, but neither S2 nor DB1 is virtualized.

Public cloud

• Everything in the public cloud is a mirror image of what is in the private cloud. The public cloud is populated by copying what is in the private cloud. An initial copy is made for each system, and updates are sent afterwards.

 A. System S1, which is virtualized needs to be copied to a pubic cloud. S1 is copied   via the link to the public cloud, unless there is a copy left over from a previous need, in which case only the differences are copied. It is converted to AMI automatically. In the case of S2, it must be copied via the link to the public cloud. Like S1, if there is not a copy left over from a previous need, it gets virtualized to run on an AMI file format.

B. System DB1 and NFS1, which are physical servers, go through the same process. They also are automatically virtualized to run on AWS/AMI.

• The two clouds are linked by the Internet or a dedicated connection via SSL.
• When any of the systems are no longer necessary, they can be disabled and deleted, or retained for future use. The copy may be retained to minimize copying time in the future.

Some high-level description continues regarding how those components work together. The actual workings are much more complex, but I have simplified them for this presentation.

Operation:

• CloudVelocity Nexus inventories all the pertinent information regarding computing power in the private cloud, including applications and supporting servers, such as file systems and databases. The configuration information is stored in a proprietary file format.
• Inventory information is passed to the CloudVelocity Cloud Manager in the public cloud. This appliance is virtualized to run on AWS (in the AMI file format) all the time. Storage and computing time for this appliance are charged per AWS pricing. The size of the appliance is negligible at several hundred kilobytes, and it does not cost much. Once Cloud Manager receives the configuration information, storage volumes for each component get allocated for each system and populated, without running it. This reduces activation time for the public cloud counterparts. EC2 charges are heavier for computing than for storage. The design is a good compromise for reducing copying time and saving on computing charges on EC2.
• Starting the systems in the public cloud typically takes three to five minutes, which is the time required to boot up a VM in the AWS cloud. They are started in parallel.
• The systems may be disabled when not needed in the public cloud. The user may expect another need for the systems sometime soon and keep a copy around, or delete it to save the storage charge by the AWS system.

Application areas:

1. Cloud fail-over: If the private cloud goes down for any reason but the operation cannot be halted, a full, earlier copy of the application systems may be started in the public cloud to take over the operation. This is called cloud fail-over and can be used for disaster recovery and for implementing features like follow-the-sun and follow-the-moon.
2. Development and testing sandboxes: More than one full copy of the application can be started simultaneously in the public cloud, while the application is still running in the private cloud. These copies are fully sandboxed and can be used for development or testing.
3. Complete move: For datacenter space constraints and other reasons, the systems in the private cloud may be cloned to the public cloud and those in the private cloud, disabled.
4. Cloudbursting: This allows extending computing power in the private cloud by enabling and cloning computing power in the public cloud, if a load surge takes place. This can be accomplished without losing data integrity in the private cloud, because two appliances can tunnel update requests back to the local site. Any changes made on the public cloud are constantly sent back to the private cloud for data consolidation, so when the load surge subsides and the copies in the public cloud are taken down, data integrity is maintained.

Patent-pending technology

Anand said that two technologies in One Hybrid Cloud Platform (OHCP) are unique, and CloudVelocity is applying for a patent for each.

The first has to do with synchronizing two data stores via two appliances that contain the inventory of computing equipment in both clouds. I will not go into detail, but according to Anand, replicating and maintaining synchronization between the two requires some work. During switchover time between the primary and the secondary copy of a VM by vMotion, pages dirtied on the primary copy are constantly sent to the secondary copy for synchronization. This requires fast (about 5 to 10 ms) communication between the primary and the secondary, but it allows a game running on one server to run continuously on another server after the move. The OHCP sends all the changes once in the form of a file and that makes it possible to send over a slower connection like the Internet with encryption (SSL). As for the moving of a running game, OHCP does not support such a feature.

The second is concerned with letting the duplicated copies of VMs in the public cloud have access over the connection to databases like LDAP in the private cloud. As noted before, because of security concerns, some servers and databases may not be duplicated in the public cloud. So VMs in the public cloud need to have access to them in the private cloud.

OHCP vs. vMotion

After discussion with Anand, I came to understand that vMotion and OHCP address different problems, but may overlap in some functionality. Both technologies move systems in execution from one cloud to another. But there is more to it. I summarized the differences in the following table.

Looking at the table above, it appears that the two technologies are not competing but can be complementary to each other. I will dig into them more in my future blogs.

By the way, I can try out their system free of charge. But wait! I am not ready. I do not have a reasonable-size private cloud myself, much less use AWS. I probably need to consult with some of my friends who are involved in Silicon Valley Cloud Center.

(Continued to part 3, which will discuss energy efficiency by cloud computing and what it means to have a hybrid cloud.)

Some people project 2013 will be the year of the cloud, and hybrid clouds are talked of as one of the trends for the year to come. See here, here, here, and many other places.

As I said before, much of hybrid cloud is just talk and not reality, and there have been several showstoppers before now.

Some of the many factors making it hard to implement hybrid clouds are mainly technical:

Technical problems

Virtual machine (VM) file format

1. Public cloud: Amazon Web Services was the first to implement a public cloud, and AWS is now the de facto standard for public cloud. It uses its own proprietary file format (Amazon Machine Image, a.k.a AMI) running virtual machines on the Xen hypervisor. Their file format is not the same as the original Xen VM format. So even if you are running Xen hypervisor for your cloud, you cannot enjoy interoperability with AWS without converting your VM’s file format. For example, Citrix virtualization environment is based on Xen, but its file format is virtual hard disk (VHD), which is also the file format for Microsoft’s virtual machine.
2. Private cloud: In the enterprise market (private cloud), VMware’s VM file format (VMDK) is the de facto standard.
3. Hybrid cloud is an attempt to use both private and public clouds to process IT demands by optimizing suitable in-house and outsourced IT infrastructures as needed. So when we want to move VMs back and forth between public and private clouds, we need translations each time we move them across the cloud boundary. It may not be very hard to do so, because there are some translation tools readily available from vendors like Amazon and VMware (vmkfstools). It may be straightforward to move VMs that are not in execution, but VMs in execution are generally hard to move with their execution state intact. See the next.

Physical movement of VMs

• If we want to exploit public and private clouds for an application in execution, that execution instance may be transported between two or more clouds to find the most suitable execution environment. One big issue is the distance between clouds. VMware’s vMotion allows you to transport your VM up to something like 100 km (80 miles) but no farther. With this physical restriction, what you can do with hybrid cloud may be limited by the distance between clouds.

Various support environment

• Cloud is not just virtualization but needs a comprehensive environment, such as management and support, including tools and security considerations. Each cloud tends to come with its own environment and idiosyncrasies, so what you can do easily in one cloud may not be as easy in another cloud. This would make managing a hybrid cloud cumbersome.

To date, most discussions on hybrid have been at a very abstract level and not at all concrete. People have talked about what we could do with hybrid cloud without referring to its concrete implementation. Recently, I came across yet another brand-new cloud company that claims to have solved the aforementioned problems. Greg Ness recently sent me email with a press release and wanted to show what CloudVelocity, his new company, is doing in the area of hybrid cloud.

I am by no means an expert in hybrid cloud computing or any kind of cloud computing, for that matter, but let me try to review how hybrid computing is implemented with their technologies. To support hybrid cloud, VMs need to move back and forth between private and public clouds. How can we implement such a move? Because an execution space is not shared between a public and a private cloud, we cannot literally move a VM across the clouds. What we do is to make a copy of a VM executing at one cloud and transport its execution status to a cloned VM at another cloud. Then we can disable the original VM and enable the cloned one. If a VM is not in execution, it is not that hard. But if it is in execution, it is much harder.

If both private and public clouds are implemented with the same technologies and the distance is less than, say, 100 km, the same VM could be transported with a utility like vMotion. But in most cases, two cloud environments are not the same (see the technical problems described above), and the distance could be greater. Also, you can move only virtualized applications but not traditionally maintained applications, because you cannot assume all the applications have been virtualized into a VM format.

We need to have carbon copies of VMs and non-VM versions of applications (that need to be virtualized) on the other side. That means you need to have carbon copies of your applications running on a public cloud. This sounds like a disaster recovery (DR) system.

Disaster recovery/fail-over system

In such a system, you duplicate the applications that are running at the primary location and operate them with options at the secondary location. These options include active-active and active-passive configurations. Active-active means that the machines (and thus applications) are live at both the primary and the secondary locations at the same time, with data being copied from the primary to the secondary sites. In this scenario, when the primary location cannot operate any longer for any reason, the secondary location can take over seamlessly. The active-passive configuration may not guarantee complete synchronization, because the passive one in the secondary location does not run until the primary location can no longer support applications.

In any event, if we duplicate the whole thing for the secondary site, as in the case of DR in an active-active fashion, the duplicated copies are always in the secondary site with dedicated servers. This situation is the farthest from cloud computing in spirit, especially for public clouds.

What we need is a solution like this:

• Copies on the other side made only when needed (on-demand).
• Noninteroperability problems overcome:
• Resolve VM file format and other incompatibilities among major cloud systems, such as AWS, Rackspace, Microsoft, and OpenShift.
• Handle physical vs. virtual applications in an IaaS cloud environment.

Now back to CloudVelocity. I visited Greg Ness and Rajeev Chawla, CEO, at their headquarters in Santa Clara. They claim to have implemented a solution to solve the problems discussed above.

From left: Rajeev Chawla (CEO) and Greg Ness (VP Marketing). See here for their bios.

They have developed a comprehensive system for implementing hybrid cloud that they call One Hybrid Cloud Platform (OHCP), which is depicted in the following picture. Applications move across the cloud boundary in five steps:

1. Host discovery—Inventory your private cloud (data center), which consists of all the pertinent IT hardware and software.
2. Blueprinting—Create a database of how the discovered components are put together.
3. Cloud provisioning—Duplicate and create VMs on the target cloud (translating VMs and virtualizing physical applications if necessary).
4. Synchronization—Synchronize VMs between the two clouds.
5. Service initiation—Let the duplicated VMs take over and disable the original VMs.

CloudVelocity’s comprehensive One Hybrid Cloud Platform.

This sounds easy. How do they do this? That will be covered in Part 2.

Long ago, I read one of his papers but this was my first time to see him on the stage. Fujitsu has cultivated working relationships with many prominent research labs and universities throughout the world. Stanford is one of them.

Everyone knows that both Stanford and UC Berkeley incubated many excellent technology companies in Silicon Valley. Many came from Stanford, like SGI, Netscape, Cisco, HP, Yahoo, Google, and Sun, and then there was TCP/IP per Vint Cerf. So Stanford must have done something right.

Dr. John Hennessy

Hennessy’s talk was about transference between universities and the real world. But this could be applied to the enterprise world as well, as in spin-outs and spin-ins. In any event, let me summarize his points, some of which I have paraphrased, and ponder on how they might be applied to the real world.

First, he said new technologies were building blocks for new industries and therefore economic growth. Those technologies came from company and university research labs. He gave examples:

· Semiconductors—initiated by Bell Labs.

· Workstations—remember Xerox Parc’s Alto Computer?

· Internet—no explanation required.

· Web—made possible by the Internet, then made the Internet even bigger.

· Biotechnology—fundamental research was done at universities.

He continued by saying that universities are a source of:

· Discontinuous innovations

· Talent

Two points he wanted to make in his talk were:

1. How to promote innovation in a university or company.

2. How to transfer innovations to real products.

He made the first point clear by saying that it is important to pick the right set of people and let them drive the direction of research. Those people should include visionaries (who have compelling ideas), explorers (who give some direction to those ideas), and uninhibited executors (who carry out the more concrete tasks and often are students.) His example of a visionary was Jim Clark, who founded both Silicon Graphics and Netscape. I had a chance to meet with Jim Clark when he founded Netscape with Marc Andreessen many years ago. Because of some interesting circumstance I will not discuss here, he gave a chalk talk just for me about how he saw the Internet and why a web browser business would grow along with the Internet’s growth by analyzing other players, the telecom and communications market at that time. It was very nicely delivered, and I felt he was a real visionary.

Also, regarding his first point, Hennessy said three things are important:

A) Live on the edge of technology.

B) Find discontinuities.

C) Find the best people.

Although there are subtle differences, I think points A) and B) are very similar in that it is important to consider what could be done even better without getting bogged down by current constraints or the way things have been designed. For A), Hennessy gave as examples both Xerox Alto, which led to the creation of workstations and PCs, and Cisco, which laid the basis for Internet infrastructures. Other examples included the Internet and Yahoo. For the second example, he listed MIPS, which he founded.

Finding good people is also an important factor. At Stanford, more than half the engineering students are from outside the US. Companies like Silicon Graphics, Yahoo, Google, Sun, and VMware were cofounded by people who came from outside this country. Those good people should be given incentives for innovation and freedom to choose their direction.

In the second point of transferring innovation, the biggest point is to transfer people but not technology. This is because:

· Inventors know the technology best.

· New technologies need advocates.

We may have known this intuitively, but he stated it clearly. Come to think of it, it makes good sense. Hennessy thinks students are good ingredients for the transfer. Get good students, grow the talent pool, and teach them how to succeed in entrepreneurship.

He then touched on entrepreneurship. Entrepreneurship alone does not work. It is also important to prepare fertile soil for startups. For this you need:

· Tolerance to smart failure

· Access to capital

· Access to engineering talent

· Access to management talent

· Access to advice

· Manufacturing capability

These requirements have been discussed in many places and in many different ways, but it was good to hear them discussed in this context. He pointed out they even apply to internal transfer.

He concluded his talk by saying that:

He did not mention cleantech, but his idea could be applied to that segment.

Cloud and virtualization components

Storage virtualization along with virtual storage and storage hypervisors have a theme of abstracting underlying physical hardware resources like server virtualization. The abstraction can be for consolidation and aggregation, or for enabling agility, flexibility, emulation and other functionality.

Storage virtualization can be implemented in different locations, in many ways with various functionality and focus. For example the abstraction can occur on a server, in an virtual or physical appliance (e.g. tin wrapped software), in a network switch or router, as well as in a storage system. The focus can be for aggregation, or data protection (HA, BC, DR, backup, replication, snapshot) on a homogeneous (all one vendor) or mixed vendor basis (heterogeneous).

Here is a link to a guest post that I recently did over at The Virtualization Practice looking at storage hypervisors, virtual storage and storage virtualization. As is the case with virtual storage, storage virtualization, storage for virtual environments, depending on your views, spheres of influence, preferences among other factors what you call a storage hypervisor will probably vary.

Additional related material:

At the ARM TechCon, I interviewed four companies that contribute to low-power computing from the perspective of various kinds of software. IS2T is based in France and provides very small footprint Java development and execution platforms. They announced two new products at the show. One is called MicroEJ and the other is MicroJvm Virtual machine . Java environments exist for enterprise (standard edition, SE) and embedded systems (micro-edition, ME). Java ME is further divided into Connected Device Configuration (CDC) and Connected Limited Device Configuration (CLDC) flavors. IS2T supports Java ME with the CLDC flavor that is suitable for very low power computing.

The spokesperson I interviewed was Regis Latawiec. Although his title is executive VP of business development, he was very technical and I enjoyed a conversation with him. Technical people tend to get enthusiastic when they meet with someone who can appreciate and understand their technologies. Well, I am not an expert but someone who knows just enough to be dangerous. In the past, I dealt with Java ME at a startup that developed and marketed security solutions for a card reader. Increasingly, cards like credit cards incorporate a smart chip that cannot be tampered with, but a card reader may not be very secure. The startup’s product was to provide a security layer, similar to the mobile information device profile (MIDP), for card readers. I am by no means a Java developer or an expert in the subject. Back to IS2T. While I was reading their press release, I had several questions and this was a good venue in which to clarify my questions. See, I am not a reporter who simply takes what was said and writes an article about it. Instead, I analyze it and express my opinion.

Regis Latawiec

MicroEJ: MicroEJ, Java development platform, is a product that consists of many elements, such as SDK, JVM, and a simulator. It uses open source Eclipse as an SDK base and expands its functionality. Because Eclipse supports multiple languages, with plug-ins, the SDK can support C and C++, which are still the dominant development languages; there are many legacy C and C++ applications still in the embedded segment.

The following is a screenshot of the SDK function of MicroEJ.

MicroJvm Virtual machine : Their JVM requires only 28 KB of flash memory and less than 1.5 KB of RAM. It also boots quickly, in 2 ms at 120 MHz. They estimate that a fully functional advanced human machine interface (HMI) requires 90 to 140 KB. By comparison, a solution from an incumbent requires 350 KB of ROM and 130 KB of RAM for CLDC.

Simulator: We can take two different approaches in development. One is to settle on a hardware platform first and then develop software for it. The other is to develop the software first and then find a suitable hardware platform. In many cases, the hardware platform is a given. Because a typical embedded system is resource constrained, we do not develop software on the target platform. Instead, we develop it on a much more powerful platform, like a PC with a faster and more powerful processor and abundant memory. But the downside of that is the need to cross-compile and download the binary to the target machine for testing. If the tests fail, we need to go back to the development environment and fix the code, then repeat the process until the code is perceived as acceptable.

With a simulator, we can develop and test software on the same platform, saving effort and the time needed to go back and forth between the target and development platforms. But it is more than effort and time; it requires extra power to do so. From the green IT perspective, this saves extra energy. Better yet, this is Java and “develop it once and run it everywhere.” So we can develop and debug software on the development platform. The very code we developed on the development platform should run on the target platform without any change. Also, no cross-compiling is necessary.

What about performance? In general, a simulator does not simulate performance, for the obvious reason. However, IS2T’s simulator provides information, such as task scheduling and memory requirements. If we have the latitude to select or adjust the target platform, we could choose the best one on the basis of the information. If we do not have that freedom, we could use the information to improve the software. Experienced programmers have a sense of how much memory is required for some particular function. If they have the memory requirement information for the current version of the software, they can use that information to fine-tune the code to reduce the memory requirement.

Legacy applications written in C and C++: C and C++ applications are dominant in the embedded world. If you provide only Java solutions and ask users to switch their programming languages to Java, you would not win them. So what IS2T did was to develop a Java platform that interfaces with those legacy applications. Java Native Interface (JNI) is a standard that allows a Java program to call programs in both C and C++ and vice versa. Their MicroJvm Virtual machine can run on either RTOS or bare metal, maximizing flexibility. Regis showed me a sophisticated diagram that was too detailed for me, so I drew the following, simpler diagram. MicroJvm Virtual machine allows the reuse of legacy programs and shortens deployment.

Support for legacy applications in C and C++

But that is not all. From the green IT perspective, this is very energy efficient. I mean each application, regardless of which programming language it is written in, consumes a lot of energy when it is developed. Software development means human activity. Programmers need to be housed somewhere, and that alone consumes energy with lighting and temperature control. Software needs to be developed on some kind of computers, and they consume energy. Source code may have to be printed out for detailed discussions, which consumes energy. Programmers may have a coffee pot for relaxation, which costs energy. If we throw them away and start from scratch in Java, the resultant system may be cleaner (in terms of logic but not of greenness) and more consistent. But additional energy must be consumed for that.

Future: The Internet of things includes many objects with networking capability. Some do not interact with humans but others do. If such an object is resource constrained but needs a human machine interface, that HMI will be very primitive because a sophisticated interface requires a certain amount of resources. With something like MicroJvm Virtual machine, many such objects could be installed with more sophisticated HMI.

Competitive edge: Finally, I asked if they had some patents to show their technical superiority. Regis’s answer was very interesting. If they apply for a patent and get it, some of the details of their technologies may be disclosed. It is their decision to keep their technologies secret. Then how do you know if they are really that good? He said he would show it by real examples. Give them a week for a fairly sophisticated specification, and they will deliver it in a week. That is proof! Also, their secrets are declared (in secret) to a lawyer organization so they can prove their invention. It is very different from what I know, but it is interesting.

In the opening speech of the 5th SVLG Data Center Energy Summit, Anne Smart, Director of Energy, Silicon Valley Leadership Group (SVLG), gave a good introduction to this topic as well as to Andrew Feldman, Corporate VP and General Manager, AMD, who delivered the keynote speech. You can watch their speeches here (a little over 16 minutes).

Andrew was CEO of SeaMicro, which developed and marketed low-power servers and was acquired by AMD earlier this year. Because the video is less than 20 minutes, you may want to invest your time in watching it yourself. But his point was simple. There is an area where we made significant progress in energy efficiency for data centers, but there is another area where we did not make much progress. The former is the facility side and the latter is the IT side. A lot has been done about energy efficiency in the facility area, such as finding better ways of cooling (hot/cold aisle and air economizer), which Andrew called low-hanging fruit. Now cooling efficiency has improved to three one-hundredths of the power required to run a server.

Now about IT. Let me inject my thoughts here before going on with Andrew’s speech. Before he started discussing the energy efficiency of IT in the data center, I knew I was going to agree with him. We have been trying to alleviate the symptom of data center energy problems by attempting to control cooling and power rather than curing the root cause of the data center energy crisis, that is, efficient ICT equipment and making better use of it, that is, running it without utilizing its full capacity. Emerson in its energy logic claims that IT is the root cause of the large consumption of power in data centers and we had better control IT. We are beginning to pay more attention to IT as we try to control data center power usage. We need to pay more attention to using more energy efficient equipment, such as SeaMicro’s, and using virtualization to increase the utilization of each server. We need more than PUE to gauge and measure IT efficiency. There have been a few data center metrics proposed, such as McKinsey’s CADE. The Japanese have proposed DPPE. Both metrics take IT’s behaviors into consideration. I recently found an interesting metric for IT energy efficiency called Par 4. I have not reviewed it yet, but the short description I read sounds interesting. I plan to write a separate blog on it.

Andrew started to talk about IT (servers) in terms of data center energy efficiency. He dealt with server energy efficiency at SeaMicro with low-power chips before he came to AMD. And he is one of the most qualified persons to discuss the subject. He bluntly stated that he was disappointed at the progress we have made in that department, because a server is the most powerful consumer of power in a data center and needs more attention.

He mentioned that like anything else energy efficiency improvement for servers began with the low-hanging fruit. In this case, it was the power supply. The power supply used to lose 18% in transit from PDU to server. Now with advances in power supply technology, it is down to 7%. But the real meat in energy efficiency is the server itself. At SeaMicro, Andrew and his team developed a server that requires only a quarter of the power necessary for other servers. But it took five years and $50M, which was not readily available from the VC community in the funding climate then. He is thankful to the Department of Energy and the State of California for their grants that made it possible for SeaMicro to complete their servers.

He wrapped up his talk by saying that we should not stop innovating in energy efficiency in both facility and server technologies. Power is such a big problem in the data center that we need to keep working on it.

Atsushi Yamanaka of Japan Data Center Council (JDCC) talked about:

• What happened to data centers at the time of the major earthquake in March 2011
• What data centers are like a year and a half after the disaster

Atsushi Yamanaka

Atsushi and I teamed up in a session last June at a different data center conference, and I think he delivered his talk even better this time. London-educated, he is an excellent speaker. His profound data center knowledge, in both technologies and business, sounded very convincing in his British accent. And no wonder, because his day job is general manager at the president’s office at IDC Frontier.

I will not recap all of his talk here but some of the information I found most interesting:

• Some rumors and hearsay about damages to data centers in Japan
• Current data center trends in Japan

After the earthquake and tsunami, rumor had it that all the major data centers had been damaged or washed away by the tsunami. Also, radiation was believed to be so bad that people were running away from Tokyo. I have been to Japan probably more than ten times since the disaster, but in Tokyo it has been business as usual. There were a few signs of the power shortage in 2011, but in 2012 I no longer see them. But don’t take my word for it; listen to what Atsushi said. He told us that data centers are concentrated in Tokyo (70%), with 20% in Osaka and the remaining 10% scattered around. Tokyo is 140 miles from the damaged nuclear reactors in the Fukushima-Daiichi power plant, far enough not to have felt any impact. Yes, the shaking was tremendous and lasted longer (especially in high-rise buildings) than most people had experienced before. But all of the data centers in Tokyo were unscratched or had minor damage, like cracked floors.

Atsushi’s presentation included a video shot during the quake that showed a guy being shaken badly as he serviced one of the racks. But all the racks had been bolted to the floor, and none of them fell down.

Atsushi also said that energy efficiency was in the blood of the Japanese people before the quake, and now they are even more concerned with saving energy, including power. Many new data centers were developed with that in mind, including those at IDC’s Kitakyusyu site, Sakura net, and Nomura Research Institute. Modularity and air economizer are becoming norms in the construction of new data centers. Those who want to find out more can ask for more information at info at jdcc dot or dot jp (change “at” and “dot” with appropirate Internet formats).

The cost of power varies drastically by location in the US, so one of the important elements in siting a new data center is power cost. The price variance is very small throughout Japan, and power quality is probably pretty uniform. The reason for the 70% concentration of data centers in Tokyo is because the city is the nation’s center of business, politics, and academia. That is changing a little, but it can accelerate still more.

We were fortunate to have with us two additional members of the JDCC:

• Jun Sato (Mitsubishi)
• Hideki Okita (Hitachi)

Atsushi and those guys compete daily but unite for the sake of the data center industry in Japan. They added some perspectives from traditional and large enterprises in Japan.

From left: Atsushi Yamanaka, Jun Sato, and Hideki Okita

The crowd at the session. After this photo was taken, more people came in, and the room was packed, with lots of people standing.

I do not remember all the questions the attendees asked, but a few stuck in my mind. Some wanted to know if the US (or Silicon Valley) is prepared for earthquakes. Hideki was polite but I am not. I do not think the Bay Area is ready. I think data centers here need more preparation.

The Silicon Valley Leadership Group (SVLG) indicated that they would like to invite the JDCC next year as well. OK, JDCC guys, next year you need to top what you just did!

Scott Jarr

When I was researching the NoSQL segment, I found it confusing enough, but there is also a NewSQL movement, which confused me further. The NoSQL movement began in an effort to accommodate the Big Data phenomenon. In the traditional database segment, ACID—atomic, consistent, isolated, and durable—is of utmost importance. The relational database was developed to guarantee ACID and for transaction-oriented applications. The traditional relational or SQL database is fine, as long as the data comes in at a reasonable speed and volume and is of limited variety. But at some point, these parameters exceeded what the traditional SQL database could handle, and new ways to cope with them were increasingly required. That is where NoSQL comes in. NoSQL, in general, does relax some of the rigid SQL rules (abandoning SQL partially or altogether and thus ACID) and accommodate these new requirements; i.e., scale-out, high availability (HA), replication, and performance. Therefore, NoSQL in general does not have SQL, relational schema, joins, or ACID (this is obvious since these are traits of the relational/SQL database). Scott put the comparison of Old SQL, NoSQL, and NewSQL on a piece of paper as we spoke. I reproduced it here. Old SQL (yet another term) refers to the traditional relational/SQL database that dominates the enterprise world.

Comparison of Old SQL and NoSQL

So in other words, in order to gain scale-out, HA, replication, and performance, NoSQL abandoned SQL/relational schema partially or altogether. What NewSQL is saying is that it can accomplish every feature in the table above while keeping the relational/SQL schema (and therefore ACID and join).

If that table is expanded with NewSQL, we have the following.

Comparison of Old SQL, NoSQL, and NewSQL

How can that be possible with NewSQL? Performance gain is a result of new architectures that remove the old baggage of OldSQL, many leverage memory for additional improvements. Actually, Gigaspaces, which I also interviewed and will write a blog about later, has an in-memory cache technology working with other NoSQL companies. However, Michael Stonebraker, CTO of VoltDB, said in one of his talks (available in a 30-minute video) that running in memory alone does not guarantee the performance gain needed to accommodate the speed at which Big Data comes in.

Mike explained in his talk that there is nothing wrong with the concept of SQL itself. It is the implementation of SQL that causes the problems shown in the table. Because of the less than perfect implementation, 96% of the time is spent on overhead and only 4% on useful work, as indicated in the following graph extracted from his presentation.

CPU cycle use in a typical SQL implementation. Most of it—96%—is used for overhead.

Unless these overheads are removed, even if all data is placed into memory, extreme performance improvement is not expected, because it only addresses the 4% but not the rest of the 96%. Typical NoSQL databases abandoned or partially supported SQL to bypass this problem. VoltDB faced the current inefficient implementation of SQL and developed their version of the SQL database from the ground up to eliminate these overheads. I am not covering each overhead in detail, but you can watch his easy-to-follow video.

OK, I get it. Then, what does this mean to the whole area of NoSQL? Does this mean the whole area of NoSQL gets consolidated into a single technology like NewSQL? Scott drew me a good figure to explain this, which he had already published in his own blog posts (the figure below came from part-1 and part-2).

From Scott Jarr’s blog. On the Y axis, data speed, size, and complexity grow upwards.

There are five areas to address in the enterprise in terms of data collection and analysis (analytics): interactive, real-time analytics, record lookup, historical analytics, and exploratory analytics.

The five areas are further explained in the following figure with applications and time scale.

Note that VoltDB is colored differently from the rest of NewSQL in the graph, but that is meant to emphasize its position in the NewSQL group. VoltDB falls into the NewSQL camp. Scott emphasized its performance superiority over others. The performance benchmark they share is 3 million transactions per second (TPS). According to Scott, the traditional RDBMS is trying to cope with the Big Data problem (velocity, volume, and variety) by scaling up (throwing in more CPU and storage power rather than using parallel computing).

In Stonebraker’s video, he said that VoltDB was five times faster than Cassandra and also faster (he did not say how much) than an unnamed incumbent’s database. When I consulted with MySQL, it was very fast (before their 5.0, which incorporated enterprise-ready features) and faster than this incumbent’s, but they could not publish the benchmark for fear of a lawsuit. I can understand that. When I consulted for JBoss, a Japanese open-source consortium compared their performance with other products like IBM’s Websphere, without any tuning. The number was not very good, mainly because those who ran the benchmarks did not know how to tune JBoss’s compared to IBM’s. After a JBoss engineer flew over there and tuned it, it improved drastically. So when we conduct a performance comparison, we need to set up a ground rule for comparison for every participant.

Of course, there are some overlaps among those technologies and their areas of applications, but this figure is a good picture of how each technology is suited for its application area. Hadoop is batch processing based and is not suitable for real-time analysis. Many people think Big Data and Hadoop are synonymous, but this clearly shows they are related but not the same. In the utilities business, a large amount of meter-read data gets collected, aggregated, and stored. By daily or monthly analysis of power usage for a particular area, a utilities company can probe into usage patterns and trends. Actually, some utilities are using Hadoop now, according to the Soft Grid conference.

Scott thinks NewSQL, NoSQL, DataWarehouse, and Hadoop will remain separate technologies because each of them is suited for some specific area of data collection and analytics. But he advocated that these areas and their tools be tightly integrated to provide analytics and thus effective real-time actions, as in the following figure. By incorporating the analytics results for long time spans into short-time analytics, more effective actions could be obtained.

Finally, he showed the current applications of VoltDB, as follows.