NEIL MUELLER: Hi, my name is
Neil Mueller with Google Cloud Platform. We're here as part of
the On-air Cloud Series. It's a series of live
webinars for Google Cloud. Today, we're hosting
two webinars-- my webinar on pricing,
and then, Sergei and team will join me after five
minutes-- five minutes when I'm finished-- to talk
about streaming analytics. Also, next month, something
to look forward to, we'll be talking about security
and IT development tools. Without further ado, let's talk
about customer-friendly GCP pricing. I've got a lot of
ground to cover today. But let me spend a few
minutes on why we're here. We're here because customers are
really excited about the cloud. But unfortunately, a
number of customers feel like they're wasting money. Now, fortunately, most of these
cloud customers are not ours. And we'll talk about
that in a moment. But according to RightScale,
which is a third party analyst firm, they wrote a report
called "The State Of The Cloud" in 2017. And they said that 45% of
their cloud spend was wasted. This is self-reported waste. So you want to believe it. On the left hand side,
you see the cause of a lot of this waste. We've got inflexible three year
VM leases that are provided by other cloud providers. And flexible VM
configurations, where you have to buy a
certain amount of RAM to get a certain amount of CPU-- also something not on our
cloud, but on every other cloud. And per-hour billing, as
opposed to what we have in here in GCP, which is
per-minute billing. So customers are feeling like
they're wasting a lot of money on the cloud at the
same time that they're saving a lot of money,
because the cloud is much less expensive than
on-prem alternatives. We're here to give you
the best of both worlds, to give you a cloud that is
exactly what you expected to pay for with
customer-friendly pricing models that you can
really easily understand. So this is a summary slide. We'll come to it at
the very end, also. If you look on the left, we've
got other cloud providers. And let's consider that
a baseline of 100%, what you would expect to pay. On the right hand
side, we've got the discounts-- the
discount innovations that we have here at GCP. And we'll talk
through each of those. So an average of 24% savings
from sustained use discounts. This is very similar
to the card that you get when you go to your
favorite sandwich shop. And if you buy 10
sandwiches, they give you one for free--
sustained use discount. When you use a lot of GCP,
you save a lot on GCP. 21% list price differences. This is just the rack
rate of our compute. If you look at
the list price, we are, on average, 20% less
than other cloud providers. And I'll talk to you a
bit more in a few minutes about why that's the case. 15% average savings from
rightsizing recommendations. We recommend within
the cloud console, if you're using too much
or too little compute, we recommend that you
can either save money, or that you should
beef up that server to make it more performant. On average, if you add
those numbers together, you get a 60%
average discount when you compare us to other clouds. That's a big discount. All right, let's talk about
all of our pricing innovations. This slide is very much a
table of contents for my talk. We're going to start
in the upper left and work our way down
to the lower right. So let's talk about
list price advantages. I've talked to you about how
we get a 21% average list price reduction compared
to other clouds. Sustained use discounts--
the lunch counter card-- up to 24% average savings. Committed use discounts,
up to 57% savings when you buy
instances in advance, when you commit to their use. Permanent billing,
up to 38% savings when you pay per minute
and not per hour. Custom machine types,
up to 19% savings. Because if you need a lot
of RAM but not a lot of CPU, you can only pay for
exactly what you want. We call that custom
machine types. Rightsizing recommendations,
15% average savings. We give you recommendations
on whether you need more CPU or more RAM. Preemptible VM instances,
we call these PVMs. And we give you an average
18% savings on workloads that can be interrupted. Network service tiers,
our newest addition-- we just launched it last week. It allows you to either pick
performance or cost savings when you buy your
network from us. Coldline, archival storage
at the speed of tape-- the speed of disk
and the cost of tape. And lastly, support. We offer you the opportunity
to pay per user per month, as opposed to a flat percentage
fee uplift on your cloud cost. Let's talk about why GCP is
the perennial price leader. Well, it has a lot to
do with our volume. Google has a lot
of infrastructure. You've probably used
YouTube and Google Search. You may have also used
Google Cloud Platform. We are a very large
technology company with a lot of infrastructure. We buy in bulk, we purpose build
much of our infrastructure. We spent more last year than
any other cloud provider-- $10.9 billion in capex in 2016. Google's vast scale means that
we can offer you more for less. We also are smart about how
we build our infrastructure. On average, a Google data
center uses 50% less energy than a typical data center. Our PUE, which is a
measure of our efficiency-- our power usage
effectiveness-- is 1.12, lower than any other
public cloud provider. This means that we use
less energy than average. If you look at the global
average of the largest data centers in the
world, they are 1.7. Now, what this means
is that about 40% of the average energy
in a data center is wasted to cooling, or
heating, or just general loss. By comparison, about 10% of
the energy in a Google center is wasted, meaning
90% of the energy that we buy goes to
providing you cloud services. We're more efficient when
we run our data centers and we can pass the
savings on to you. We're also more
efficient, because we deploy very interesting
technologies like machine learning. You may have seen
the article written about DeepMind, who applied
some of their technology to our data center. And if you look on
this graph right here, time is on the x-axis. On the y-axis is PUE, this
power use effectiveness ratio that I told you about. The closer you are
to 1, the better. Google is closer to 1 than
other average data centers. And when we deployed DeepMine's
AI in our data center, it reduced cooling by 40%. So you can imagine what
this ML looks like. Let's take, for example,
we've got, like, 10 dials. And each dial has
10 positions on it. That's 10 to the 10th
power, or 10 million. That's a lot of instances
to calculate through, which is exactly what DeepMind
is calculating through, and finding the exact
right middle ground where we can safely operate
our data centers, and also simultaneously
reduce cooling. This allowed us to reduce
the cooling bill by 40%. Were not the only one noticing. Greenpeace gave us an A. Other
cloud providers in the space didn't get an A. We are in
this as a result of energy transparency, committing to
being 100% renewable by 2017. And our green
advocacy that we do. We think our data centers are
beautiful inside and outside. And in fact, this is one
of our murals that we have. If you send me a Tweet and
you're the first person to send it to me, I'll send
you a $25 Starbucks gift card. This is part of
our mutual project. We paint many of our
data centers with murals. Just send me the state
name via Twitter, and I'll send you
that gift card. This is a very
large data center. All right, let's talk about
sustained use discounts. This is the sandwich car
that I talked to you about. So on the x-axis here, we have
the monthly average usage. And so if you use 100% of
this VM for the entire month, you can get a 30% savings. We're seeing our
customers experiencing an average of 24% average
savings, because many of our customers use that
VM for the entire month. In other words, they're getting
a lot of free sandwiches. What sustained use provides
is if you use us more, we will give you a lower price. This is automatic. There's no upfront
payments required. There's no lock in, and there's
no complex decision making. You don't even have to remember
the sandwich card for us to give you the discount. So it really is just that easy. You might be
wondering, we use terms like customer-friendly pricing. But are we really
that friendly when we're doing the calculation
of sustained use discounts? I can tell you, yes we are. If you look at the
graph on the lower left, you can imagine in
the actual use case where you've got a
whole bunch of VMs and a whole bunch
of projects across your entire organization,
what we could do is we could not be
friendly about it. And we could calculate
your actual usage and say that you hadn't used
a single VM the entire month. But that's not how
we calculate it. What we do is we
have this technique that we call inferred
instances, where if you look in the middle,
where we group them, we group all of the
instance usage together. And then, on the far right,
we compute a discount based on that. So we put them
together in a way that looks like a very successful
"Tetris" game, which gives you the maximum possible
sustained use discount, truly customer friendly pricing. Let's talk about CUD,
Committed Use Discount. What this is, is when you
buy cores and memory in bulk. You can either select to use
a one year or a three year commitment. And you can change your
machine types at any time. If you do this with us for
one year for three years, we'll give you up
to a 57% discount-- big discount. So what you're doing here is
you're buying scores and memory in bulk. If you buy it for one
year, it's a 37% savings. If you buy it for three
years, it's a 57% savings. If you compare this to other
similar offerings in the cloud space-- for example, AWS's
reserved instances-- this is, again, a report from
a company called RightScale. They're a third party
advisory firm out there. And this is a report that they
did very recently in March, when we released
Committed Use Discounts. We are the blue block. In this case, lower is
better, because we're talking about price and costs. So because of the
rigidity of the AWS RIs, and the flexibility of the
committed use discounts, the realized benefit
for you is much higher. In this case, the
costs are much lower. Let's get into a bit more detail
about Committed Use Discounts. Like I said, these are
available in one and three year. You commit to a
certain number of vCPUs in a certain amount of memory. This is valid for any
Google Cloud Compute Engine, or GKE non-shared core VMs. And it's ideal for predictable
steady state workloads. I said that it's a committed
use on CPU and also memory. There are very flexible
limits on the ratio between memory and CPU. So anything from 0.9 gigabits
of RAM up to 6.5 gigabytes of RAM per CPU-- inclusive those two numbers-- is an appropriate ratio. So if you've got a very
memory-intensive application that doesn't require much
CPU, committed use discounts is still going to
be good for you. The discounts apply to
aggregate regional usage. For example, of your usage in
Iowa, or Oregon, or London. This is a summary slide talking
about Committed Use Discounts. What I want to highlight here
is the very far right block, which says that in order to
attain these deep discounts without upfront payment, you
can attain these deep discounts without upfront payments at
all, because we don't require you to do any upfront payments. This is very different from
other reserved instance-type structures that you
see there in the cloud. Per-minute billing. So, on GCP, you only
pay for what you use. So for example, if you're
using a VM for only 11 minutes, you're going to pay for 11
minutes, not 60 minutes. This is a big differentiator
with from other cloud providers. Let's walk through
a short example. So on the right here,
you see that if you're using the virtual machine for
between 12 and 25 minutes, your average savings
is up to 38%. Custom machine types. If you go with another
cloud provider, you will be dictated a certain
amount of CPU and memory. And if you need more
CPU or less CPU, you have to pay for the
stuff that you don't use. So right here on
the x-axis, you've got a series of machines. And the gray would be waste. In this case, on GCP, the
waste would be much, much less. On average, we're seeing
customers save up to 19% when they use custom
machine types. This is a screenshot that I
took from the user interface. We've got these dials where you
can dial up the amount of cores or the amount of
memory that you want. And we did the math,
and it turns out that there are 695
different compute types. If you compare that to
other cloud providers, we have 110 more compute
types than they do. So you get exactly what
you want and nothing more. Rightsizing recommendations. This is very similar to
a custom machine types. What this looks like
in the user interface is a recommendation
where it says, hey, you're using
a bit more instance compute than you have purchased. Would you like to upgrade? Or in more cases than that, you
have purchased an instance type that is larger than you need
and you're paying for something that you're not using. Would you like to save
yourself a little money? And all you have to do is
click the button right here to either dismiss all of
those recommendations, or to learn more. And with one click,
you can activate them. It's the opposite
of requiring you to plan in advance for demand
that you can't possibly forecast. 25% of monitored instances get
a recommendation on average. So that's a lot. It'll be a good portion
of your instances. Now, you can very easily
dismiss these or take action. Recommendations take a
look at CPU and memory. On average, we're seeing 60% of
those recommendations offering to save you money,
and 40% offering to increase your instance size,
suggest that you get a beefier instance, because you're getting
more traffic than you then you may have expected. If you use stack
driver, the granularity of these recommendations
improves quite a bit. Let's talk about
preemptible machine types. Preemptible machine types
are very good for workloads such as batch or scientific,
predictable workloads that you don't think
are going to go down. They're exactly
like regular VMs, except for they are
80% less expensive. They can run for a maximum
of 24 hours, at which time, they'll be shut down. You can terminate them
earlier if you like. And it offers
predictable billing compared to other
cloud providers that have similar
structures like this, but they're based
on marketplaces-- where you would have
to hire somebody to go into that marketplace
and understand the flow price of that machine instance. Whereas here, it's just a
flat 80% off, so much easier to forecast. We view this as
customer-friendly pricing. Like I said, they're very good
for various kinds of workloads that you know might be
able to be interrupted. For example, batch
jobs, pipeline builds, after hours processing,
big data workloads, analytics, media and coding, scientific
work, financial modeling, dev test, crawling,
that kind of workloads. What is most amazing about PVMs
is how easy they are to deploy. I mentioned with other
types of structures like this from other
cloud providers, you might need to hire
somebody to go in there and understand the
marketplace dynamics of their interruptible
instances. Not with ours. They are a flat 80%
price reduction. And you can either
activate it using this code right here that I've got on
the screen, or in the cloud console with a point and click. Customers love these. For example, Mark Johnson,
the founder of Descartes Labs, has been able to predict
corn yield more accurately than the United States
Department of Agriculture, all using PVMs. So he's benefiting from
GCP at 80% the cost and producing incredible
output for his customers. Network service tiers. Last week, we became the
first major public cloud to offer a tiered cloud network. What this means for you
is that you can either choose to pick performance,
which we call the Premium Tier-- think premium gasoline. If you pick the Premium Tier,
it is on average, 70% faster than other cloud networks. It is 14% to 80% more expensive
than other competing clouds. And so the ratio of
18% more for 70% more power, that's a very good ratio. We feel like it's a good
value for your money. If you pick Premium, you're
riding on the Google Network, which is a non-circuitous
direct private dedicated path from your cloud workloads
to your end customers, egressing out of a POP very
near to your end customers. If you choose Standard,
you'll benefit from a 6% to 9% savings over other
competing clouds. Standard network uses
the public internet, which is very similar to
what other public clouds will provide you. This is often not a direct path. It suffers from lower
security, lower liability, but it's a very good
value for the money. Let's talk about storage. On this graph, on
the x-axis, you see the retrieval frequency. So on the left, you've got
Glacier, an AWS product, that offers retrieval frequency
of three to five hours. It is very close to the same
price as GCS Coldline, which offers millisecond access time. We also offer Nearline,
Regional, and GCS multi-regional. The reason we've put their
storage options next to ours is what you'll notice
is ours are always below theirs,
which means that we are less expensive for the
retrieval time required. So we think that
our storage is going to be a better value
for online data, whether it's hot or
cold, whether you've got millisecond access
time that you need globally reliable, or less. Coldline is a highly
available, affordable solution for backup, archival,
and disaster recovery. It really is incredible. We offer you the speed of
disk at the cost of tape. But you might be
asking yourself, how am I going to get all
my data up into the cloud? Well, for that, we just launched
something about last month called the Transfer Appliance. This is a physical
appliance that we rent you for a window of time. You load your data onto it,
and then you send it back to us using a courier. The capacity for this is
up to 1 petabit compressed. It's the largest transfer
appliance in the cloud. Because a lot of our customers
are moving a lot of data to us, and we wanted to give
them as much capacity as we possibly could. The Transfer Appliance
is all about speed. In this animation,
what you're seeing is how fast it would take
a Transfer Appliance to be shipped out to you,
for you to fill it up, and for you to
send it back to us, and us to load it into
Google Cloud Storage. On average, 43 days. If you compare that over
a typical network, which is 100 megabits
per second, it is going to take a lot
longer than 43 days. As you can see, it will
take multiple years-- like more than three years. Let's talk about what
we're doing for support. Support hasn't evolved. What we are doing to
evolve it is we're offering role-based support. What this means is that you
pay a flat price per user, per month, as opposed
to a percentage fee uplift on your cloud usage. So you might be accustomed to
paying a certain percentage, like 10%, to 20%, to 30%
uplift on your cloud usage. We are evolving that to this
new method, where you pay a flat fee per user per month. So let's take an example. Let's say that
you're a development shop with 10
engineers, that would be 10 engineers multiplied
by $100 per user per month, multiplied by 12. So that's $12,000
in annual support for this development shop, which
is likely much, much less-- and certainly, much,
much more predictable than if you were to pay a flat
percentage fee on your cloud bill. It also is a lot closer to
getting a value for your money because we're charging
you per engineer, as opposed to charging
you a percentage uplift. The old model, we
call account-based. The new model, we
call role-based. Account-based, you're accustomed
to long-term contracts, standardized tiers,
and like I said, percentage-based uplift
on your cloud bill. With role-based, you've got
a very flexible environment that you can change month to
month, no commitments, very customizable configurations. You tell us how
many users you want. And it's very predictable,
because you know how many users you're buying. This is coming very
soon, and we're incredibly excited about it. It offers development support,
production support, and also support for your
business critical apps. All right, let's talk
about some key takeaways. If we start in
the upper left, we talked about why Google was
a perennial price leader. We operate at a vast scale,
and we pass those benefits on to you. On average, we're 20% less than
other cloud providers, just the rack rate. Sustained use discounts--
just like when you go into a sandwich shop. This is up to 24%
average savings. Committed use discounts. If you are able to plan
ahead and know that you're going to use us, we don't
make you pay upfront, but we offer you up to 57%
savings for a three year contract. Per-minute billing. Up to 38% savings by
paying by the minute, as opposed to
paying by the hour. The analogy I gave, the example
I gave was using 11 minutes and paying for 60,
very frustrating. We don't consider that
customer-friendly pricing. And so we don't make you do it. Custom machine types. Up to 19% average savings when
you pick any configuration of CPU and RAM. Rightsizing recommendations. 15% average savings
by auto scaling compute around your needs. We give you these
recommendations. 60% of the time,
they're to reduce the size of the workload,
offering you a savings 40% of the time, on average. They're free to beef up your
instance and save you downtime. Preemptible VM instances. Up to 80% savings-- and this is a flat fee,
not based on a marketplace. Up to 80% savings on workloads
that can be interrupted, like batch jobs, or
crawling, or scientific work. Network service tiers,
our newest member to this list, where you can pick
performance and get 70% higher performance than the
average cloud provider. Or savings, and save 9%
compared to other clouds. Coldline, archival storage. And the Transfer
Appliance, which allows you to move lots of
data to the cloud very quickly. Support. We are evolving support
and very excited about it. With role-based
support, customers pay a flat monthly fee
per user, per engineer. Here's that graph again. Let's talk about
where we've come. So another cloud
provider on the left will charge you a
baseline of 100%. You can experience an
average of 60% savings for compute workloads on GCP. 24% is coming from sustained use
discounts, 21% from list price discounts, 15% from
rightsizing recommendations-- big benefits for you. Customers are picking
JCP, in part because of our customer-friendly
pricing, but also for a lot
of other reasons. On the left, we offer you
a true cloud economics. That's all about the pricing. And we've walked through
a number of examples where we provide you
very transparent, friendly pricing
structures that you can-- without taking a lot of time-- take advantage of. Most of these you
take advantage of just by using the system at all. You don't have to plan ahead. We offer you future-proof
infrastructure, where you can scale
your business smoothly and responsibly. For example, we're
the first cloud to offer Silverlake Intel chips. And you didn't have
to plan for that. You didn't have to buy
them a couple of years ago. You can buy them the
second that you want them. And then, the second you
stop wanting to use them, you can stop paying for them-- big advantages that you
get from using the cloud. You can also get access
to incredible innovation. Google is known for our
artificial intelligence and machine learning. We give you access
to many of our APIs, including Vision API, Natural
Language API, BigQuery. You can use these
to really spark off a lot of innovation
at your company. And Google grade security. We offer multiple
layers of security so that you can control your
cloud from chip to chiller. OK, with that, I
think I've been told that we've got a
couple of questions that I probably
should help answer. OK, I'm back. I can see on my
screen here that we've got five questions to answer. The first one is-- where did
the idea for sustained use discounts come from? It predates me here at Google. I came up with the
lunch counter example, because I eat sandwiches. And when I buy 10 of them,
I get a free sandwich. And it sounds a lot like
sustained use discounts. The big difference between
sustained use discounts and the lunch counter
is that I'm always forgetting my little
card and losing it. And that never happens, because
we keep track of it for you. Question 2. Are GCP pricing innovations
only for compute? Many of them are for compute. Rightsizing recommendations,
sustained use discount, committed use discount. Those are for compute. However, many of them are not. For example, Coldline-- where
we offer you the speed of disk at the cost of tape-- is very innovative and is
a storage innovation, not a computer innovation. Network service tiers also
is the opportunity for you to experience 70% faster
networks than the average cloud provider, or 9% lower costs. And that's a
networking innovation. Question number 3. Can I change my instance
sizes and still receive committed use discounts? Yes. Unlike other cloud
providers that offer you committed use discount
structures that are inflexible, our are flexible. So you buy a certain
amount of CPU and a certain amount of memory. And within that
bundle, that basket, you can change exactly the
kind of VM configurations that you want from
minute to minute, day to day within that month,
within that year-- whether it's a one year or three year
committed use discount. Question 4. What is the difference
between SUDs and CUDs? Sustained use
discount, let's see if I can think of
a funny analogy. Sustained use discount is when
you go to a lunch counter, and they give you
one of those cards. And if you eat 10
of those sandwiches, they give you a free sandwich. Committed use discounts is
when you go to the car wash and you say, I want
to buy 10 washes. I'm going to buy them right now. And then, I'm going to
use them over the next one year or three year. And as a result, you get a
cheaper rate per car wash. That analogy falls
apart a little bit, because we don't make you pay
ahead, and the car wash would. But other than that,
it's a decent analogy. Number 5. Will Rightsizing Recommendations
support all operating system compute instance types, not
just Linux but Windows servers as well? Yes. It supports all of the
instances that we have. I use it on mine. And it gives me recommendations. For some reason, for
mine, it's always telling me to make them smaller. It never tells me
to make them bigger. Maybe it's because I don't run
that many popular websites. Hopefully, you do. And it works on all of
mine, Linux and Windows. Are there any more
questions, guys? OK, thank you so
much for your time. Please stick around. Our next webinar is on streaming
analytics, very exciting. Sergei and team will be here
to talk you through that. Next month, also, we've got
another on-air webinar series by Google Cloud. It's on security and
IT development tools. OK, for now, thank you
so much for your help.