Open source guru Marten Mickos dissects Amazon outage

In this week's episode of Cloud Cover TV, open source chief Marten Mickos sits down with Jo Maitland to analyze the aftershock of the recent Amazon cloud outage.

We discuss:

  • The Amazon power outage and its effect on Eucalyptus
  • Why we shouldn't be surprised by outages, especially with massive servers
  • Amazon's communication style (or lack thereof)
  • The possibility of cross-multiple clouds with Eucalyptus' software
  • Private cloud's growth with Eucalyptus
  • How cloud installation reduces costs and complexities
  • Eucalyptus' strategy for hanging in the cloud game

Read the full transcript from this video below:

Open source guru Marten Mickos dissects Amazon outage

Jo Maitland: Hello, and welcome to Cloud Cover TV, our weekly show
on all the juiciest news in the cloud computing market. I am Jo Maitland,
here in San Francisco, and this week I have Marten Mickos with me, he
is the CEO and Founder of Eucalyptus. Welcome to the show, Marten.

Marten Mickos: Thank you, Jo. Happy to be here.

Jo Maitland: Just to kick off this quick interview, Eucalyptus makes software
that basically lets companies roll their own infrastructures of service in-house.
One of the interesting features of it is that it seamlessly hooks up without
modifying any code to the Amazon cloud, to AWS. I am curious, given the
crazy news this week of the outage, it was three, four days, still, customers
this week are having some problems with services running on AWS, whether
that feature, that ability to seamlessly connect and move workloads to the
Amazon Cloud is still a feature people are going to be interested in?

Marten Mickos: I definitely think so. Eucalyptus was designed as a very
powerful cloud software platform, as an engine, you could say. Then our
founders decided, I was not the founder, I joined a year ago, but the founders,
the six PHDs decided to build on top of it the same API as AWS uses, the
same API as Amazon has. This was to build a complement, an on-premise
complement to what they have on the public site, so if you run something on
AWS, you can move it over to the on-premise side, on Eucalyptus. If you develop
on Eucalyptus, you can go live on Amazon, you can move workloads back
and forth, and you can build hybrid solutions on top of them because the API is
exactly the same, it is the same platform for the developers.

Last week, Amazon had this famous outage which really harmed some of the
web companies who were using AWS and were using it based on just one
region. I would say, in the grand scheme of things, we shouldn’t be surprised
that things that human beings create sometimes break down, specifically,
when you have such a massive service as Amazon. Even if you have just one
defect in a million, one defect in a billion, or even in 10 billion, you quickly get
up to 10 billion, in terms of volume, and you will have those outages. It was
very bad for some of the companies, and I think there are lots of things for
Amazon to learn and do better next time. I also think that it will not stop the
growth of cloud computing, I do not think it will harm Amazon's business in the
long run.

Jo Maitland: Do you think there is something wrong with the design principles
of their cloud, or is this just a case of large numbers and so much scale,
therefore, there’s aggregated risk because it is just such a big platform?

Marten Mickos: I think they have show that the concepts they have chosen
and the designs are very sound and very robust, but something was wrong
in the implementation, because the malfunction, the outage actually spread
beyond availability zones, which they are not supposed to do, so something
went wrong there. Some people say that the biggest mistake here was that
Amazon did not communicate early enough and directly enough about what
was happening, and that is actually how they could have avoided some of the
problems, by just letting their customers know what was happening.

Jo Maitland: Why are they not doing that? They have been criticized for
this in the past, and their communication is still radio silent on this one.

Marten Mickos: I don't know. I think they have built a fantastic cloud. I think
they are an amazing online book store, and I read books on the Kindle
every day, so I really admire the company. No company is perfect, and
maybe this is an area in which they would want to develop.

Jo Maitland: There is also an option to build across multiple clouds. Is
that something you that can do from the Eucalyptus software? Could I
work with Rackspace? Could I work with the other public clouds, as well
as AWS, from your internal software or . . .

Marten Mickos: You certain can, in terms of the design, we built a product
that can support any cloud API we would like to support. Today, Amazon's
API is 60% of the market, or more, so we have chosen to focus on that,
because that is where all the action is. Unfortunately, we have not seen
many other clouds yet really grow into prominence in this market, so that
is why, just for business reasons, we are squarely focused on being
compatible with Amazon applications and we do that, not just so that you can
move applications, but also so that the skill level you achieve on one is
useful on the other. Say you read a book about Amazon's Cloud, you have
essentially read a book about Eucalyptus. Say you develop a tool that runs
on Amazon, it will run on Eucalyptus. A tool that is built for Eucalyptus will
function on Amazon. It’s a way of lowering the hurdle of getting going with
the cloud, by using the de facto standard, which today is API.

Jo Maitland: Right. Although it does not have you spread your risk across
multiple clouds.

Marten Mickos: That is true, but you can establish a Eucalyptus on-premise
cloud, where you move, that you have as a backup cloud, if you like. I am sure
that in the next few years we will see other public clouds with the same API,
then you will be able to move workloads and have them as backup for each other.
Sometimes we say that it isn’t a cloud if it has strict boundaries, clouds need to
be open to interaction and clouds bump into each other; that is the analogy of the
cloud. It is very malleable.

Jo Maitland: In terms of private clouds and people running private clouds on
Eucalyptus software, are we talking about 50 companies, 100, 300? Where are
we in the adoption?

Marten Mickos: Every time people start using Eucalyptus, unless you stop it from
doing it, it will phone home and check for new images. We have measured
25,000 such starts; it means that, in the world, at least 25,000 Eucalyptus
clouds have started up. Some, of course, are experimental clouds and short-lived,
but many of them are production clouds all over the world. Specifically, we see
attraction with web properties, typically big brand owners who operate web
applications or mobile applications, we see a lot of government agencies, we see
academic institutions, universities, and we see just start-ups and enthusiasts
who love cloud technology and would love to roll their own, so they download
Eucalyptus and they install it. You can install it on even a single laptop, it is not
really a big cloud yet, but you can do it. You can have 5 servers, 50 servers,
500, 5000.  

Jo Maitland: How many of those are moving workloads to AWS?

Marten Mickos: I wish I knew.  

Jo Maitland: 5%?  

Marten Mickos: You know this is open source software, so they are under no
obligation to tell us what they are doing. I would say there is a reasonably
good overlap between Eucalyptus users and Amazon users.  

Jo Maitland: I am curious. From what I have read and the CIOs that I have
talked to that are building private clouds or even IT directors that are moving
in that direction, they have this sort of notion of, "OK. We are just going to
spin up a private cloud over here just to get the CEO or the CIO off our
back. We can say we got one." It was a bit like the ‘90s; ‘Yes, we got the
internet.’ They are spinning up this silo now, and adding yet another silo
of IT infrastructure that is deployed and managed differently than everything
else that they have. Is that ultimately not just adding to the complexity and the
cost of everything, when the whole goal of cloud is actually to try and reduce
the cost and reduce complexity?

Marten Mickos: I think the opposite. In the long term it will reduce
complexity because that little cloud installation will expand, they will see
the benefits of it, and they will expand it into more and more areas of
their data center. I do think it is smart of CIOs to start small and test it out,
because not only is it new software that you need to understand, it is also
a new way of dealing with software, so managing your data resources and
your IT resources is different. More than before, you need to have the system
admins work together with the application developers, traditionally they have
been silo’d, but with cloud computing they are coming more together and
they work together.

Jo Maitland: That seems to be one of the big barriers too, the political
boundaries inside of companies.

Marten Mickos: It is always the barrier. When you come with fantastic new
technology and you can demonstrate amazing cost savings or increases in
utilization, it still is the human being who is the slowest moving part in the
equation. People have built their own careers, they have certified themselves
for certain technologies, they are just not interested in bringing in a new one,
until you find those pioneering spirits who will do it, no matter what. Those are
exactly the small clouds that you talked about.

Many times when you talk about the cloud, we think about something absolutely
massive like Amazon's cloud. In reality, most clouds are very small experimental
clouds, today, with tens or hundreds of servers maximum. That is where we have
enormous penetration with Eucalyptus, in companies, governments, and start-ups.  

Jo Maitland: Equally, though, that means that they could just switch these small
clouds off. I am wondering what that means for the survival of your company and
how long you are going to have to be in the game for. It seems like this shift in IT,
people compare it to the shift towards client server and we are talking years, and
years, and years. How do you manage that?

Marten Mickos: That’s the good part here. Big shifts always take many, many
years to happen, and you have to have the endurance and the patience to wait it
out. It completely follows the technology adoption lifecycle that Geoff Moore
described in 'Crossing the Chasm' in the ‘80s. I think he wrote the book in the
‘80s, and it is completely true today. It will take time, but we have a huge update
for our technology among the pioneers. At the same time, the world has grown so
much, it is not just the US and Europe anymore. It is Brazil, Russia, India, and China;
massive markets that we are working with.  

For instance, Eucalyptus, our first office and sales office outside of the US was in
China; that is where we went. We felt it was the most attractive market for us.
Even if you could argue that it will take time, and how do you know that they do
not close those and shut down the cloud? Sure, many will do that, but there are
even more who are not doing it; who are really moving forward and seeing the
strategic benefit of the cloud.

Jo Maitland: Marten, thank you for being on the show.

Marten Mickos: Thank you, Jo.  

Jo Maitland: This has been Cloud Cover TV. Thank you for watching, and
tune in next week for more insider news and interviews in the cloud computing

View All Videos

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.