David Karp is the founder
CEO of Tumblr and Kevin
is the CEO and founder
of instigram and this
panel is on hockey
stick growth and both
of these founders have experienced
hockey stick growth at different stages.
Instagram is a much younger
company and Tumblr is
just taking over the web.
What do you, you do like 250 billion impressions a day?
250 million a day.
250 million, I'm sorry, okay.
About 7 billion a month now.
No 250 million, yeah.
Yeah, sorry.
So you are 250 Million a
day and you were
at 250 million a month
just a few months ago, right?
Yeah, the sort of exponential
growth we started to see is
really sort of counter intuitive like
to think back a year ago
how many servers we were ordering.
We were thinking about scale it
just doesn't apply at all today
and looking forward the next
six months, planning is
an interesting exercise, but often feels futile.
Right.
I think we have some charts, if we can put them on the screen now would be great.
So that's your pretty,
I dunno its hard to tell,
I mean it's been
pretty, what were the inflection points?
If you have to look at that, and
pick as many inflection
points as you want, I mean
what's the story that that chart tells you?
So, one reason
it's been hard to, I dunno,
the interesting thing the
whole way through is if you slice
that chart anywhere and zoom
in, it looks pretty much the
same at any level along the
way it's just that, you know,
today the numbers that we're talking
about, I think are a bit more
substantial than they were, you
know, two years ago when even
though the graph looked pretty much the
same, slope hasn't really, you
know, it's always looked like
it was tickingout its never
been easy to
think about, as far
as reflection points, the only
thing I think that's really
pivoted over the last four
years is been whether or
not most of that growth was international versus domestic.
And our first year was almost entirely US traffic.
That next year we started
to see significant international growth
for the first time, to the point
by the end of the year, by the
end of our second year we were
something like 75% international traffic.
Then it started to even out
and now you can see in that
graph, orange is US domestic traffic.
That makes up about 40 percent of our traffic today.
So it's now starting to, I
think, much of the
significant growth we've seen over
the last six months has been,
sort of mainstream adoption overseas.
So Brazil, Southeast Asia,
Japan, various European
countries starting to take to
Tumblr the way the US was
starting to adopt Tumblr a couple of years ago.
So now lets look at the in instagram chart.
That's a different scale, what is that scale?
Actually how many photos are posted per minute?
Per minute?
Absolutely.
So we are getting about ten photos per second right now.
And that was, last
time I talked to you, was six or four?
It was a few months ago.
I mean, I think a lot
of what you just said really
resonates with us as well
which is, you never expect
growth to keep growing the
way it has been growing until
you wake up one morning and
you say, that growth we
saw last month, that we
thought was really large, just pales
in comparison to what we're seeing today.
And it's really exciting
to wake up every morning and see that.
I think on this chart, you were asking about inflection points.
t's been a pretty steady
growth since Christmas of this past year.
Christmas day was like this
huge day when everyone
opened their iPhones and started
using camera apps again because
the iPhone 4 had this better camera.
yours is international versus US.
It's interesting because I actually
think that our app was
international from the very beginning
and that actually surprised us.
In fact, it was
probably more international at the beginning than it is today.
I think most of our growth early on was in Japan.
And, only now am I
seeing large growth in the United States as well.
Do you think there's also a difference
between mobile and web,
and there's a lot of mobile
components at Tumblr, obviously, but,
if you think back to the,
you know, where you were
at this stage, right, I mean, is there any advice you have for Kevin?
Yeah, Kevin seems
to be a few steps ahead of me actually.
I think he's been hanging out with the right people.
One thing that, most of
the stuff that I look back and
kinda go man I could've done better
was, sort of team
building stuff, or more than
team, really company building stuff,
that it's just, it's easy
to start setting up
when it doesn't really matter if
you kind of fail or need to change things dramatically.
You know if you're 5 people or
a dozen people it's much easier
year to kind of say, okay,
this is the scaffolding for the
bigger company that we're going to grow into.
Rather than, now we really
doubled the size of our team
in the last 6 months, now we're
thirty plus people and I'm
holding all team meetings for the
first time in our four and a half year history.
You know, I could have started practicing
those back when we were a dozen people.
Right.
So, as your usage
grows, right?
Obviously you have the scale
of the technology and that's a big challenge in and of itself.
I'd like to hear you guys talk a little about that.
And how many server
melt downs have you lived through,
if you can still count them.
But then on the other hand
what you talk to here which is
not only to scale technology
you've got scale of the organization, right?
And how do
you do that, and does that
come up faster than
you thought it would, would
be you need to fill certain
roles that you didn't
think you'd have to fill for
6 months and all of a sudden you really need it.
So,I'm terribly curious to hear
how Kevin's been approaching the what
their attitude toward the scaling and the architecture is.
I'll tell you that
the thing that our investors
kind of beat into me for
years before it finally
clicked, and I sort of
accepted or understood it, was
it was really just a question that
I would get at every board meeting which
is, "Is the way that you're thinking
about scaling, or the
way you are thinking about technology going to scale?"
What I now
see what they meant was, whatever
position you're in and whatever
expectations for where your product
is going, exponential growth
is sort of unfathomable until
you cross over to
the other side, and all
of these steps along the
way require completely different solutions.
As a scrappy team of two
engineers at Tumblr for three
years, we wanted to
come up with clever solutions and
engineer our way out of
any scaling challenges which came along.
Realistically, negotiating transit, building
up multiple data centers and
building an application that can, live
and write to multiple data centers
are things that are really
sort of impossibly large challenges
for people that haven't been through them before.
And approaching it
that way of, there
are just things that, this is the
reason that Facebook has very
smartly gone through several, sort of,
iterations of their engineering team, bringing
in people who have been through these
things before and know how
to think about this next level
of scale and it's why
at this point I completely defer
to our operation's and engineering's
teams who have seen the
stuff before, cause it's not stuff
where I can help with
anymore and it's not a place where I can be clever anymore.
How are you in
terms of all of
these companies that just take
off, whether it was Twitter or Tumblr.
Right?
You have these periods where it's
down, and everyone makes a
big deal about it because it's not available.
And if people start to rely on
a service, it becomes noticed when it's not available.
It's been a few weeks since the last time I heard that Tumblr was down.
Where are you?
I would say first of all, when
that stuff happens its absolutely heartbreaking,
I mean it's like the most gut-wrenching experience
in the world and I don't
think absolutely gut-wrenching experience.
Well lets see,
what took, it
took me too long to realize that
for us it was actually that
we needed more resources, we needed
to be in more data centers, we needed a bigger engineering team.
And interestingly, if you
actually zoom out and look at
our all last-team meeting,
we looked at that traffic graph next
to our up-time graph for the last year.
And you can see nine
months ago, when it started getting
a little bit rocky where we had,
you know, fantastic up-time and performance
for years before that, all of
the sudden, we were hit with
this growth that, again, it
was that exponential growth that's, just
sort of, impossible to imagine
where you're, you know, you're just, like,
OK we're ordering, you know, how
ever many servers a week and
in a few months we'll be ordering this many servers a week.
You don't realize that, like, you
know, it's a lot of work
for your managed data centers to set up racks of servers.
They might be fine to turn on
one server for you a day
but all of a sudden you place
an order for 10 racks and
they're like, okay give us a couple of weeks.
And you know, it It's a
completely different world and a completely different set of solutions.
Anyway, the graph
that we looked at was watching
our up-time get kind of
rocky over a few months,
and then our down-time in
December, we had a major outage
that took our site down for a day, which was like the most painful thing in the world.
And, it happened that
in December was when we started
to do our real push for hiring,
recruiting really spectacular engineers.
We quadrupled the size
of our engineering team since December,
and if you look at our up-time
graphs since then it just
straight slope or its
actually very steep slope straight up
and out or back to several nines up time.
And you have great 404 error.
Getting better, yeah.
The thing that was really important
to us through this whole experience was
that, or has been
really important to us, is making
sure that that's not something that kind of defines our company.
I think it was sort of
a stigma that hasn't quite left Twitter.
I think all of these stigmas you can leave behind.
I think generally it takes somewhere
between three months and a
year before people...The internet has
a very short attention span and very short memory.
Have you had any big meltdowns Kevin?
Yeah, we had actually the first
day, we launched and we
had no idea we we're going
to be as big as we became.
And that first day we were
on a single computer in
a managed data center.
And I still remember my
co-founder Mike, calling them up
and saying, "We need another computer,"
and they're like, "It will be 6 hours."
And we didn't half six hours,
so we ended up moving
to Amazon actually for everything
and I think it's a trend now with
a lot of the smaller start-ups including
ourselves to move to
the cloud because of the flexibility that you have with growth.
On the downside, you're paying
more than you would if you owned your own assets.
But that flexibility I think really allows you to scale.
At least at our level, very quickly.
And then...
But then when
Amazon's down...
Yeah, when
Amazon's, so it's, you know,
but at the same time if you
do Amazon correctly you don't
face those things and I
think a lot of people were
in one availability zone and that really bit everyone in the butt so.
Let's talk a little bit about,
this is a high quality problem to have obviously.
Right?
It's a great problem.
Can you manufacture growth, or
do you just build the best
product you can and what happens happens?
I'm not sure you can
manufacture growth, actually you can, but I'm not sure it will be sustained.
Like you see this in the app store all the time.
We live and die by you
know the top three apps right,
and you see apps that
have a one-star ratings pop to
the top and you realize that
something is going on there, so
you can manufacture short-term growth
but I don't think that long-term growth,
the nice growth you want
to see in a products, you can
manufacture outside of
building a great product and outside
of building a great team that
builds that great product.
When you build something that people really
No comments:
Post a Comment