Ben Horowitz was working at America Online as a vice president and
general manager when his old friend Marc Andreessen called and asked
him to join his new project: a data center automation software
company called LoudCloud.
Now the company is known as Opsware and has marquee contracts with such
organizations as systems integrator EDS, cable giant Comcast, financial
services firm Lehman Brokers, the U.S. Department of Energy, and telecom
player NTT Com.
Internetnews.com sat down with Horowitz to talk about his
company’s crusade to make the data center a friendlier place to work and how
to use grassroots efforts to the company’s advantage.
Q: You and company chairman Marc Andreessen had great success with
Netscape. Any predictions about Opsware? Will it become that big, that fast?
I certainly think that we are going to do it again with Opsware. I
would say we are not there yet and we are working very hard.
We found a problem that absolutely needs solving.
Q: What is the story you are bringing to the market?
One of the most important points we try to make out in the market is that
there is a trend in computing in the form of a shift from client-based
computing to Web-based computing. And it’s interesting in that a lot of
people don’t acknowledge it as a platform shift. That has led to a lot of
complexity in IT.
The big change when you go to the Web is that it gets radically cheaper
if you develop an application — like 100-fold cheaper. And so because of
that, the number of applications has proliferated. As a result, the
number of servers has proliferated quite intensely. The comparison is this:
500,000 servers shipped in 1995 versus 5 million shipped this year.
The other very interesting thing for us is that in 1996, you paid $3 in
hardware for every $1 in labor or service, and the price for service was
high. Today it is reversed. It is $3 in labor for every $1 in hardware cost.
As a result, labor has just skyrocketed. IDC recently said it costs the
industry $95 billion to manage these servers. And that creates an
opportunity for Opsware.
Q: What about your relationship with EDS? Is the new contract just an
extension or are there changes on the horizon?
We just signed the renewal. It was originally a $52 million software
contract, and then we signed a new $50 million contract. We’re working with
them to automate their entire environment, which consists of 65,000 servers
and 154 data centers. If you think about the scale of what they are doing,
it is pretty unmatched in the marketplace other than maybe IBM. But given
the way that IBM is organized, it would be very hard for them to automate
that whole thing.
As for changes, we’ve continued to evolve. One of the things we’ve
done together is co-authored DCML, which is the Data Center Markup Language
standard. This is not only necessary for the industry but also for the
relationship. They have a lot of other systems and vendors that they
want to make sure that, when they are putting these products in on those
other systems, they plug right into Opsware.
So we needed a standard way of moving information across that network,
which is why we are asking for other vendors to comply.
Q: Speaking of which, how is the specification doing? Given the lack
of support from IBM, Sun and HP, it looks like you are fighting
an uphill battle.
We’ve had a lot of experience with standards in the company actually
coming from Netscape. [Our Chief Technology Officer] Tim Howes co-invented
LDAP and was very instrumental with SSL, HTTP, HTML and the coming
standards. What we found can be summed up from what we heard in a quote from
an ITEF member, where he says, “Committees are great, but we will go with a
rough consensus and working code.” [laughs]
If you think about the history of standards, the ones that have worked
have had a rough consensus to them, and the ones that haven’t have had the
big guys in it. For example, consider the failed standards — X500, Corba,
X400, OMG — these are the ones with the biggest names in computing. In the
case of HTML, Microsoft fought us to the death on that one. Remember BBML?
That has been pretty consistent. The same with LDAP and X500. LDAP was a
bunch of guys from the University of Michigan, and then Netscape began using
it. And the last guys in were the big companies. It makes a lot of sense for
them to be last in because the big value in the largest corporations is
integration.
You pay them a large services fee and they will integrate all of the
products you buy, or you can buy a proprietary solution from them and they
will integrate their product line. So if you now say that there are going to
be standards, then that really levels the playing field. The larger
companies fight the standard or muddy the standard.
Q: Do you see any of the largest vendors breaking ranks and joining
your vision?
I think that they will break ranks just as soon as there is a range of
operable implementations. We already have some pretty big companies behind
us: CA, Mercury Interactive and BEA. As the implementations come out from
those vendors, and as you can buy Opsware and Mercury, and you
can plug them together without professional services, people are going to
like that. And they will tell the larger companies that they will not buy
something from giant software vendors unless you’ve got the same interface.
Q: What kinds of requests are you getting from customers?
Linux and especially SUSE Linux is getting traction. We are getting
requests for VMware, which we will eventually support. We are seeing
servelets do a little better than the J2EE and [Enterprise Java Beans]
programming model interestingly. And this is another case of simplicity
trumping complexity.
Q: Are there any other technologies on the horizon for Opsware?
Open source programming languages are developing at a really rapid rate.
Python came out of nowhere and has a huge developer following. Ruby seems
like it is getting a lot of traction. And then there is Eclipse, which is an
incredible developer tool.
Q: I know your core business is focused on data centers, but what
about something like an Opsware Light for smaller businesses?
I think that an Opsware Light is definitely a potential interest for us.
Our current product is really designed for a minimum of 100 to 200 servers. Once you get
past that, things get complicated. It is like air-traffic control.
If you have two planes and they are doing fine, you don’t need automated
air-traffic software.
