Transcript slides

All Change – why the economics of
Cloud will make you think differently
about Java
OR
-Xmx: £100
https://www.flickr.com/photos/teegardin/
Important Disclaimers
THE INFORMATION CONTAINED IN THIS PRESENTATION IS PROVIDED FOR INFORMATIONAL
PURPOSES ONLY.
WHILST EFFORTS WERE MADE TO VERIFY THE COMPLETENESS AND ACCURACY OF THE
INFORMATION CONTAINED IN THIS PRESENTATION, IT IS PROVIDED “AS IS”, WITHOUT
WARRANTY OF ANY KIND, EXPRESS OR IMPLIED.
ALL PERFORMANCE DATA INCLUDED IN THIS PRESENTATION HAVE BEEN GATHERED IN A
CONTROLLED ENVIRONMENT. YOUR OWN TEST RESULTS MAY VARY BASED ON
HARDWARE, SOFTWARE OR INFRASTRUCTURE DIFFERENCES.
ALL DATA INCLUDED IN THIS PRESENTATION ARE MEANT TO BE USED ONLY AS A GUIDE.
IN ADDITION, THE INFORMATION CONTAINED IN THIS PRESENTATION IS BASED ON IBM’S
CURRENT PRODUCT PLANS AND STRATEGY, WHICH ARE SUBJECT TO CHANGE BY IBM,
WITHOUT NOTICE.
IBM AND ITS AFFILIATED COMPANIES SHALL NOT BE RESPONSIBLE FOR ANY DAMAGES
ARISING OUT OF THE USE OF, OR OTHERWISE RELATED TO, THIS PRESENTATION OR ANY
OTHER DOCUMENTATION.
NOTHING CONTAINED IN THIS PRESENTATION IS INTENDED TO, OR SHALL HAVE THE
EFFECT OF:
- CREATING ANY WARRANT OR REPRESENTATION FROM IBM, ITS AFFILIATED COMPANIES
OR ITS OR THEIR SUPPLIERS AND/OR LICENSORS
2
Your Guides for Today’s Journey
Steve Poole – IBM
Making Java Real Since Version 0.9
DevOps Practitioner (whatever that means!)
@spoole167
Chris Bailey- IBM
technical architect for runtime
monitoring and diagnostics
@chris__bailey
This talk is intended to make you
• Think about how you use compute resource today
• Think about how that use may need to change
tomorrow
• Starting measuring your application profile now
• Try out cloud technologies and gather real experience
• Wonder where Java is going next
This talk is about how this sort of
measurement:
GB/hr
Is already changing your life & the direction of the
Java ecosystem
The ‘Cloud’ has a lot to answer for
Outline
• Part 1 – The economics of Cloud provisioning
• Part 2 - How Java measures up
• Part 3 – The API economy and what that
aaaaaaameans for Java
• Part 4 – Our thoughts
Part 1 – The economics of Cloud provisioning
Why ‘Cloud’ ?
Your business wants needs to get
ideas into production faster,
cheaper and with less risk
A big challenge is its traditional
infrastructure
A local, hand-crafted, static
environment which requires inhouse specialist support, doesn’t
scale well and requires long term
investment and commitment
https://www.flickr.com/photos/sylvar/
What ‘Cloud’
promises
a virtual, dynamic
environment which
maximizes use, is infinitely
scalable, always available
and needs minimal upfront
investment or commitment
Take your code – host it on
someone else's machine and pay
only for the resource you use for
the time you use it
AND be able to do that very quickly
and repeatedly in parallel
How quickly do you need to get good code into
production?
• Would you believe < 1hr?
• Case Study: A fashion retailer can show
measureable increase in sales if a item similar to
that seen in the media can be placed on their online store landing page within 1 hr of it appearing in
public.
• Each product placement is different so they need a
fast, agile, approach that does not jeopardize their
on-line stores availability and quality.
• We know how to do this..
Cloud computing is real.
Major vendors are
providing substantial
capacity and it’s growing all
the time
Businesses see the opportunities
Improved value for money,
decreased time-to-market,
shorter time to value
“I can now get my ideas into
production in hours,days or
weeks. I can get immediate
feedback AND then I can
improve the idea and repeat”
Hybrid Cloud is coming to a data centre near you
http://www.fodey.com/generators/newspaper/snippet.asp
We’re moving to a more
industrial scale
Why buy one computer
for a year when you can
hire 365 computers for a
https://www.flickr.com/photos/skohlmann/
The ability to have ‘cloud
burst’ capacity is
changing the way
software is being
designed, developed
and supported
“Compute on demand” – it’s what we always wanted
https://www.flickr.com/photos/vuhung/
Cloud Economics
We really are getting closer all the time to
‘Compute on Tap’
https://www.flickr.com/photos/leunix/
Cloud Economics
But with taps come meters…
https://www.flickr.com/photos/beigephotos/
Cloud computing: compute ==
money
Money changes everything
With a measureable and direct
relationship between $£€¥ and
CPU/RAM, disk etc the financial
success or failure of a project is
even easier to see
And that means…
Even more focus on value for
money.
Someone
will be
looking at
your leaky
app
American Society of Civil Engineers
Part 2 - How Java measures up
Compute ==
money
$ == GB/hr
-Xmx: £100
Easier than ever
before
a business can buy
a CPU
Just for how long
they need it.
No long term capital
investment.
Just as much as
they need
Real costs – it’s confusing
Offering
RAM Cost
CPUs
IBM Bluemix (CF)
$24.15 GB/Month
4vCPUs per instance
IBM Bluemix (Containers)
$ 9.94 GB/Month
4vCPUs per GB
run.pivotal.io
$21.60 GB/Month
4vCPUs per instance
Heroku (Hobby)
$14.00 GB/Month
1 "CPU share" per 512MB
in an instance
Heroku (Professional)
$50.00 GB/Month
1 "CPU share" per 512MB
in an instance
Amazon EC2 (SLES)
$16.56 GB/Month
1 vCPU per 4GB in an
instance.
Example costs
Bluemix (CF)
Amazon (ECS)
Running a 512mb
container for a year
$289.80 for 4
cpus
198.72 for 1 cpu
Reducing memory
footprint by 10% saves
$28.98
$0
Running 10 containers
in parallel
2890.80 for 40
cpus
$1987.20 for 10
cpus
Reducing memory
footprint by 10% saves
289.80
$0
N-Body Memory Footprints (lower is better)
N-Body Completion Time (lower is better):
N-Body Completion Time, normalized for
100MB of memory (lower is better):
Unnecessary baggage
(you have loads)
Java 9 modularity will help but
you have to consider footprint
across the board.
Choose your dependencies
wisely
Your choice of OS & distribution
is important.
The aim is ‘carry on only’
Your application isn’t going on a
long trip
https://www.flickr.com/photos/armydre2008/
Java applications have to get
lighter.
Startup times
How long do you want to wait?
How long do you have to wait?
If the unit of deployment and scaling is
an instance of a service it needs to start
FAST
BTW – think about this:
Everything that happens at startup –
happens every time, all the time.
https://www.flickr.com/photos/91295117@N08/
Do you need to preemptively start
instances ‘just in case’ due to start up
time?
To bad – that costs
Java & fast startup time – It’s known
for it!
Application developers can reduce service
startup time by deferring optional costs to
when its needed.
Maybe even create services with different
behaviors rather than one with optional
behavior
But it’s not enough
The JVM needs to revisit all the places
where startup time was traded for
throughput and turn them around.
what about
“ Everything that happens at startup –
happens every time, all the time”
For container based services – all this
start up effort happens multiple times
during development and testing (let
along during production)
https://www.flickr.com/photos/quinnanya/
The rub
We don’t have a good way to capture
all this effort or formalise starting a
JVM from a precanned image.
(Shared classes doesn’t hack it)
Other languages have better / faster
startup!
https://www.flickr.com/photos/dno1967b/
And it’s always the same result.
AND you will pay £ for it every time
Runtime
costs
• Q: How much RAM does
your application use?
• A: Too much
https://www.flickr.com/photos/isherwoodchris/
Runtime costs
Most cloud providers will charge you for your RAM usage over
time:
$GB/hr. (Sometimes the charge is $0)
Increasing –Xmx directly effects cost. Something businesses
can understand
Net effect – you’ll be tuning your application
to fit into specific RAM sizes.
Smaller than you use today.
You need to measure where the storage goes.
You’ll be picking some components based on memory usage
increasing the amount of memory for 1 service
increases the bill by the number of concurrent
instances
https://www.flickr.com/photos/erix/
Multiple languages on the
JVM.
Nashorn JavaScript engine
What’s the benefit of
running them on the JVM
vs having a native service?
They can take more
memory, and take longer to
execute.
Cloud applications are
increasingly
heterogeneous.
Anyway they share data
not objects
delivered in JDK8
Utilizes new JVM level
features for performance
Avatar.js provides Node.js
support on Nashorn
Results of “Octane”
JavaScript benchmark
using Java 8 pre-u20
Node.js is 4.8x faster
Avatar.js is >10x larger
Feb 12th, 2015: Avatar is “put on hold”
https://blogs.oracle.com/theaquarium/entry/project_avatar_
update
Simply
Java applications are going to be
running in a remote, constrained
and metered environment
Whether your application is large or
small, granular or monolithic.
Someone will be paying for each
unit used
That person will want to get the
most out of that investment
https://www.flickr.com/photos/rvoegtli/
There will be precise limits on how
much disk, CPU, RAM, Bandwidth an
application can use and for how
long
Part 3 – The API economy
And what that means for Java
The API economy
If your company has data it will
eventually be shared and monetised
Really.
Cloud APIs are one of the fastest growing areas in our
industry.
Sharing data and services though APIs is enabling new
opportunities and solutions
Everyone is getting into the game.
What makes a good cloud api
?
• roughly in selection order.
vailability
100% of course with performance SLAs
elievability – Are those published 100% metrics true?
ost – how much and what’s the unit of measure?
iagnosability – can users debug problems without you?
xcitement – is there a vibrant community using the
API?
unctionality – what else can the API do?
Where you code runs day-to-day and momentto-moment will be driven by economics, legal
requirements and how much risk your business
wants to take.
Your code has to scale better, be more
efficient, resilient, secure and work in
constrained environments
You will have to design, code, deliver, support
and debug code in new ways
It’s going to be scary
Almost
everything
about your
application is
effected
design, coding, deployment ,
startup, execution, scaling
debugging, security, resilience …
https://www.flickr.com/photos/mjtmail/
How
scary?
Resilient
applications
“Everything in the cloud
fails all the time” : Werner
Vogels
Design for short term failure: something fails all the time. Expect data and
service outages regularly
Fail and recover: don’t diagnose problems in running systems. Kill it and
move on
Every IO operation you perform may fail – do as few as possible
Every IO operation may stall – costing you GB/hrs and resources– timeout
everything quickly
Every piece of data you receive may be badly formed – check everything
Retry, compensation, backout strategies– these are your new friends
Debugging
Remote support for your family?
Fancy having to do that for your own
apps?
You have to assume:
You will never be able to attach a
remote debugger to a failing app
Ever.
All problems must be resolved by
local reproduction or logs and dumps
https://www.flickr.com/photos/carbonnyc/
You will never be able to log into a
remote server.
It gets more challenging.
Failures during deployment
or initial startup can be difficult or
impossible to diagnose.
If your service instance didn’t start
there is is little chance of logs
being kept!
Learn to love logs, dumps and
traces.
Remote log stores and tools are
going to be your best friend
BTW: they’ll cost too
https://www.flickr.com/photos/hinkelstone/
Debugging
Debugging
• Q: Why can’t you just keep the failed
instance around?
• A: You can – if you accept the £££
consequences…
Hard metrics and
limits
keeping a failed app
around or having apps on
standby can be costly in
multiple ways
Runtime costs and taking
up vital resource allocation
Part 5 – our thoughts
It’s all change
How you design, code, deploy,
debug, support etc will be effected
by the metrics and limits imposed
on you.
The JVM and Java applications
have to get leaner and meaner
You have to learn new techniques
and tools
https://www.flickr.com/photos/beigephotos/
Financial metrics and limits always
change behavior. It also creates
opportunity
More thoughts
• Do we need a JVM anymore? If your container has code that will
ONLY run on one OS/arch do we need hardware abstraction like
class files and bytecode?
• Linker coming in Java 9 helps reduce footprint and some
startup time.
•
We need more AOT to convert Java into executable code once
only
• Individual service lifetimes are short so dynamic recompliation
is not useful unless the generated code is shared. How do we
share compiled code cheaper than it costs to generate the
code?
• Remember – you’ll be paying for all the ‘wasted’ CPU / RAM
etc.
Has this talk made you
Think about how you use compute resource today?
Think about how that use may need to change
tomorrow?
Want to measure your application profile now?
Want to try out cloud technologies and gather real
experience?
Made you wonder where Java is going next?
https://www.flickr.com/photos/pasukaru76/
You stay in wonderland and see
how deep the Java goes
The story ends – you wake up and
Java is what’s it’s always been
Thank you