Applications in the cloud

so let's keep talking about scaling up to the club have you were actually
running
application developers running wrapped in the cloud how many have you are
fitted mens
K out what are the rest to the group Timothy how many of your manager who
know nothing about how dinner here for education
okay for the record you're just alright of Khan because
speak up I won't buy de about Robert did you raise your hand if it is a minority
really insulted by I'm not a sysadmin
yeah I know how right exactly you're too good for sysadmin
alright so the whole point of this is about talking about
I'm for from the developer perspective okay but that feed into dev options that
have been because
in the new peaceful world rely and lay down with lambs these two groups should
be working together to really build a successful app
so the agenda is learn a little bit up platform as a Service on the prime not
and then a little scaling and that that currently in 30 minutes
so the assumptions is I present sum up what we know and then we discuss like I
really want this to be a discussion session
there in the small obsession and these areas are new enough where I would like
to actually get be back for people in the crowd and get a healthy debate going
good I'm gonna bring up a term I'm gonna prep you for it now
so you can read nor the rest my talking to get the argument in your head ready
micro services that like stirred debate among anybody hear anybody excited about
that
nobody yet thank you I'm caught I saw your hand got some are calling you when
we get to it
the depth of scaling out running about a great scaling picture would be
and I couldn't really good one but I thought it was a great picture in the
end
because the scaling is like when the web with that where farewell to member that
when they had problems with scaling
but off with cum like Moby Dick coming up under your book and taking you out
when something really big hit you and you haven't built for web-scale that was
on the fly analogy so there's this guy to scaling
the first is vertical which is bigger stronger faster
okay so camera guy just want to say though where the camera he left the room
it's gonna be hard for speakers to walk back and forth this way cuz theres like
all
obstacle course like American Ninja rate here I'm bigger stronger faster
and that horizontal scaling a swarm a beast swarm right for this to the
example a vertical scaling
right the traffic is looking a little white boy
right and he you got some pretty scaling about happening and what you do is you
build something bigger stronger
after overwhelm the traffic add more RAM and at a fair and you start adding all
for the really fun stuff
I need a bigger server right and i'm looking you
the average age in this room is somewhere about minor older
so most in your YouTube this version of scaling right
when from that when traffic you to cum in you call the opportunity by more
servers now
or I need a bigger thurber through more RAM in it right that right
okay thank you that was a lukewarm at best buy yet that mode what most about
it and that's how most of us learned a builder applications right
you build one big job after one big PHP app
and then if you are good scaling you put your database server on another server
right for you have an absolute ever and a database server needed but minus the
machine three had independent than that was your
you know I'm web-scale I'm this
is vertical to horizontal scaling
right is the idea like a former vans so that when something big company need to
handle it you just start throwing irreplaceable
the replaceable things in there no such thing as a replacement
irreplaceable really means it irregardless that but not real word
you keep going and at it right small though and the carried out largely an
large spider not one amp by itself in paris by to keep doing a bunch
and if one gets bitten by the spiders not a big deal because there's a whole
bunch more
and surround pick up a load right so
why did it take so long for us to get to this the first part we needed cheap
commodity hardware
and I would say that relict cheap easy to buy
commodity hardware that was my cow to back up your Linux Magazine
right where you could miss temple your own server but like we needed cheap
commodity hard when I think this was actually
we need more slow to catch up I think this is actually the biggest reason we
didn't have
vertical scaling what Moore's Law in a broad sense what made
0 hardware yet although I don't wipe well because
cloud my Berto I think it's why I put that thing for catching something I
didn't think too much about
I'm and then there's even creation and deletion
so this is where like the people at Chapin poppin all the other day about
people come in
is the either creating and deleting servers I remember when we used to after
servers
it would take weeks in and provisioning server would take a very long time
and that is you can't do that if you're going to vertically or horizontally
scale
and then after those two things are in place you can start to build software
grow sideways
right and I think that was a big paradigm shift that happen for a lot of
people rather than I'm gonna build it bigger and
look for optimization in my tight loops is there to say I'm gonna look for ways
I can to keep adding servers in
and its gonna grow across the side
two different kinda propaganda talk right now I'm going to move out of the
camera but it's not propaganda at all actually I'm only putting us up at the
way most people scale
this this is how we still an open shift but don't pay attention to the
particular word
the idea is this could be infrastructure as a service to their body no
infrastructure the services
the difference between platform as a service infrastructure as a Service if
you don't shake your head
saying yes that means I will have to explain it so does everybody understand
a word with him
platform in the service and Infrastructure as a Service no thank you
infrastructure services Amazon is a
OpenStack at CloudStack right you get virtual machines on the fly but then
it's your job to do what you want with their servers
nothing else a provision Platform as a Service is Cloud Foundry OpenShift a
Roku
we say I want this stack proficient at all for me and then you have to pick up
more of the
application tier they need it I never touched networking or anything else
taken up that took did that's a model though applies whether it pop over the
servers are
infrastructure the service he basically partition a buncha stuff and then you
have something that handle spinning up more stuff
right and so here we can actually spin across here which is the same thing you
would have to build yourself an infrastructure-as-a-service and I'm not
saying one is better or worse
but what i'm saying is you basically start breaking your things up into
different here's your load balancers on one here
your application stuff in a different here they can scale independently and it
built to scale independently
and your database on another tier okay so this get the students were
I 23 minutes left I need to slow down
because we're out I'm I'm this get this too I think we're we're are today
which is micro services right so does everybody in the know what micro server
sitters
now okay how many you know about outsell
every everybody in the room know what so it stands for service-oriented
architecture
right so service-oriented architecture big Jace thing
am a basic and action say dotnet as well
it was basically web services with a capital W in the capital F
ran the idea there is you're sending XML back and forth with altered the
complicated protocol
but you had this thing called a service but in the middle and that would
coordinate all the transactions between all the different pieces
and then what came after up
web services with a cupcake capital W and the capital S what was the next big
thing that was actually built for humans
that handle that same kind of stuff rest thank you
see that one the hard and good job see I'll give you positive reinforcement if
you take a risk
I'm rest with the thing that came out to replace
for humans web services with a capital W like WS star
right and so what this starts to enabling us to start to use the web
to build those same kinda service-oriented architectures
without all the I need to spend sixty million dollars on Oracle training in
WebSphere and all that other stuff in
have all these com- pany several levels Enterprise Architect figuring out how
that's going to work together
and the micro services in I think the term was coined by power but I don't
know for sure
it's the next evolution is sewer they had a very bad job not acknowledging
that so it's kind of its roots it is the router bit
but it is done more to human scale and the idea with my curse
micro services there's was your typical old school application right
everything bundled together into one huge monolithic ap and when you need to
scale it you need to scale up the whole application at
over and over again which is quite complicated and can be quite expensive
because it is purple thing is that the limiting factor
and everything else under loaded you're actually scaling up a whole bunch of
other stuff that you don't actually need
right that old-school so with single monolithic applications
micro services did you raise your hand before in the back
whose good poster child for micro services hoop
doctor no no with the poster company that use micro services in production
thank you who said that thank you in the back Netflix is the poster child for
micro services right to the idea with micro services is
with another one that there's been a lot a write-up about lately and it's one of
my favorite music services I'll evangelize even though I don't work for
them
yes Spotify rightful spot about organizes like
Mitt Pena belong to each wanna meet
right the team is not this
the team is this the team there's another team for this
and there's another team for this and those pieces written relief
independently on their own schedules
and on their own servers so even within one company this could be node
this could be job up and it can be jobber and the database team you have
three different kinda databases it whatever the team needs
and what they do is they communicate ugg across
the using rest and the communique use proper usually within a min
I & Event Manager like a message queue right
instead of some service oriented architecture bus and so this is what
about if I does they have teams that work like that
and they really each piece independently and then they don't say there's no
specific hardware requirement for what that have to run on
it can run on any pieces or whatever the devil op sys admins
spin up right and they work with the department has a min notice look-see
nothing the same on any boxes
is complete independence this and
gets taken out I can spin up the pieces on the ant anywhere else I want to
I don't have to say all right that on that specialize HPU
no fence HP but that HPU actor a iraq software and only runs on their
you can even state even spins on rail I'll pick myself for it
you have to be on that rail box with the following configuration at No
there is a box that is our box we have we have three sizes in those boxes and
I'm gonna have to spin it up on a large
from a nap this business up on a small ran so that would get us to scaling in
the cloud
there's some advantages to this and I'm gonna ask you to start naming what for
many advantages are
because everybody's about the fall asleep in a row in the first session
so what it what could you see if it manages to myth architecture
going to start your got good get
right its object-oriented at the 15 min level
where the systems level you decoupled everything
right you disagree on interface which is your ETA like I'm gonna send you this
URL
you're gonna give me back is Jason and that's all you have to agree on as long
as that agreed on
is totally decouple I don't have to sit here and worry what the release schedule
at the Red Bull if I'm on the other team I need to talk to the red one
I don't have to worry about their release schedule and at the time I
really schedule with them
right my release schedule becomes my reschedule their early schedule their
release schedule
as long as everybody obey the same contract
which is object-oriented in libraries right so take a good concert in building
up but you can also bring about that
when I come to the bad parts you already answered one thing I try to get more
competition
what another banner Jeff
exactly so the team get the pic what they think is the best technology for
their part the stack
right so I'm in charge the part which is publishing a bunch of data out really
quickly
I don't want you to come to me and say you got a bill that and JE
right I want note and I wanna be able to hire no developer going to be doing real
a think right off the bat on my JBoss partners are gonna get really upset with
me that I just about but still
if the new cool kids are doing a thing stuff wanna do it would note in Mongo
right and so you can hire note in Munger developers without having to have a
complete
back you do have to build the team though but yeah
next
electrode together I'm supposed to repeat everything and you said it very
nicely
but because we're recording a happy kinda say it again I'm
you get encapsulation this ap if I bring in some library
look like they were now back in Java ME motor mostly programming PHP
Java or run job in your shops
I there any other languages other I
a funny roomie I Python
okay so you'd this guy's install a pep
up a python dependency right and named Paul a different version I collide with
this group back in here
what happens you're screwed and you have a month everybody arguing with each
other about which had the burden they're going to use and why can't we do this
and then
this group in that oh my god we had this dependency particularly on that really
old version
but with you're happy I I think all the speakers are gonna move this way dude
no difficult obstacle course over there
okay so what over here I can install whatever pipeline dependency I what
and I don't care in my team have to decide but its only 19
right so there's a lot less configuration where were yelling at each
other about who gets the gap what version
I mean in a real if you're dealing at the infrastructure as a service level
it could be Ubuntu this could be Red Hat this could be done net
right is not even at the library level you get encapsulation even for the
operating system level
and some would argue and this is where I take of my micro services preaching at
good I don't really necessarily agree with this is that we should all have
their own databases
right I think sometimes there is value in having them share data bases you lose
somebody encapsulation good you're talking to the same data source
but sometimes it makes sense to have like if this is the user management
system in this is a reward system I kinda don't want to maintain to set the
user's
so I want them talking to the same database maybe
okay we've got them really good reason why with Deanna everybody that %uh
Infrastructure as a Service of like at the system level throughout this is a
good idea
like from you this admins I'm expecting you to say something at some point yeah
I'm not look up put together a yeah if it had been looking into your email
or met you know you're looking at that pic sums are your monitor your sis was
making sure they're staying up
up what let out what are some really good reason for sys admins about this
about going to this horizontal scaling yeah yeah
the vision thing cost you can take pick this one to be that they did that node
server
you can do this one would like by 12 make the RAM right
this is the JavaServer you gonna do that we like eight gigs RAM
right so you don't have to with this one
this one has 64 gigs RAM does it have to support everything
and you don't you only this been up as many abuses you need for that piece
right yeah right
so the I I bet pirate talk just so he said it to provide monitoring because
yours
monitoring earth a simpler set up things I bet if we were to talk to the Spotify
sysadmin there or maybe they actually said in the article and I didn't see it
I bet they have 15 min to find each application right so
if you have a bunch in because 10 things I remember with this team is that they
hated about the
developers bringing in new technology if this is four different technologies in
here
their monitoring team the whole team has to know all four different technology
that once
and when things go wrong you're having a beer all for those out
here if I'm the the sysadmin the sign to the you
word about people assigned to the use your application
I just have to know a code that in the user application
and it did know that I lived in a hundred no run node and run noted its
best
yet so failover
it I think they'll ever become usually what's happening inside here than the
band bus is the talking between them
right so you talk across an event but the same messages
out on the event person whoever able to subscribe answer can answer
which is better than in this case recall tightly bound inside
it's also easier to scale up if you start to see that work
to can't handle failover if you to die all avenues been up two more v
I don't have to spin up to more of these
just to get those two little boxes right there yep
right to his point I is that it
in an emergency recovery you need to start is at a premium this
you need to think of this and then you need to think a death many need to get
this
you have to come up with a plan for each one other yes I agree you have to have
more granular planned
but some would argue that actually easier than trying to think about in
certain ways than this
with another drawback now since I'm now down to 12 minutes
yep
yet
I would agree about that though his point as security or things which are
overarching across all your sweet tooth applications are harder to think about
and harder architect right work on Dec
because I've been trying to build a micro servers myself now like I'm
actually trying to really relive the pain
rather than just eat the stuff in them I got the coolest so good
you should drink some I'm and that is one of the first things I ran into
like how do you handle authentic like to the user authenticate against the app
here
how does this app this app this app this app and this app all know that you heard
of any hidden allowed to act is that service
as opposed to hear that's right you have to
and then but then you also have to get into token exchanges and you have to get
into
00 often felt like you have to get at some level yea
which is more complicated than here right if I did it here
I would just tell the app server to have domain level authentication and I could
blow up
it bear for me right so there are pieces that you have to give up like to me
like application wide services are much more
big more thought to be built right rather than relying on your
app server or whenever you're using to give you that out of the box
right that one of the drawbacks coordinated large helpings arm
more difficult and chances for you screw up that's right so the net the other one
would have to pay more attention to the wire
right the network actually starts the matter a lot more
yet
right
to yet this is and which so the point was that they can hear on the video
in the traditionally weak architected for things networks for me this
not for that Ryan here the who all
this and just one pipe this way right from application sitting about here
and the other thing what do that mean like what back in our days when we were
building apt
what was told to us was the most expensive
big start
round trip yeah so back in the day you
like the reason you what are the reasons you didn't do scaleable would but by
putting your database server on another machine
where you were told it is expensive to leave your machine
and go to the database machine and then wait for that to come back right
so there are certain use cases where this makes more sense
if I really care about the time it takes for the date like
a microsecond leveled response from the database this is not going to be the
most efficient
nor help me make my goal right because there's going to be at least a couple
round trips prolly to get anything done
so that when the drawback to micro services yeah yup
that no I don't know that I think that's always a good nano
so micro service so the arrest the far end the ones you think like
rest have to be exactly like the thesis right we have you know
it have to be obey the whole entire structure I don't believe that about
rest either
I think that rest I use it more more pragmatic perspective
and from this time are actually more pragmatic about that either I think it
kinda silly to say everything has to have its own database
but there that would help with latency I may not be talking as many hops
but I actually think that freaks me out I'm not ready to take that we be able
all your data
yeah yeah
so you know that the that the new early adopter promise your gonna run into if
you try to do this right now right
a almost all over for the last however many years
have been taught to do this and you going henner higher
architect at this point most in are going to how to do this and if you going
higher sys admins for the most part they're gonna have to do this
and most developers are gonna know how to do that right
and so if you'd I think getting are
re human people more all humans are
people I'm getting people I was tryna I was trying to avoid the word resource
breakers what does that mean what it if you refer to your people is resources
they get to refer to you as overhead has met with the right
so I'm not using the word resource
II but then I couldn't think fast enough about a
and stop the jet lag from flying around a lot I'm it harder to find good staff
they can do that now right so I think someone is going to be actually you
taking the burn on yourself and training up your staff
by like taking the like do some small test project
there's a bunch a great articles out there about people talking about taking
their monolithic
I did a really good one about taking a monolithic Ruby application my and then
walking through all the steps to get to this and it doesn't show much for be
code so I was
not that I'm comfortable with it yeah yeah
no i i think so you're playing the same it he's talking a lot
the German and black shirt where the blue black and then if a row
yup a it be a the
he's talking a higher level so i think i agree with you that it
hiring at this level much easier hiring at somebody who can see this picture
and manage that entire picture much harder because there's not as many
people
out there who've done this and have done it well
so you have to put try to put people who like Netflix and Google and
good luck with that right because that hard cock too much or a good start up
yeah right
so the high school student in the crowd said you can always hire how it goes
didn't
I'm peak he'll be taking lb has resume in the back corner after my talk is done
I'm though yes few net of it you can also take
40 something year old guys and gals and teach them as well
I'm but might the only well was not going to be about hiring high school
students right now
that is a great idea I'll leave it at that it has both pluses and minuses
yeah
yes I want it so it he said I making the scope it a micro service to turning what
the actual scope with my core service can be challenging
and that's one of the things I got into an architect my app like how
fine-grained I want the service to be
right like do I want one service that's just for reading users
and another service that's just for writing users like I should be
completely different different services
or do I have a user service is that too much so like
some people vision we have an internal lift the Red Hat
called core which is all over Java Debs if you want to talk about play more
so we have everybody on our team that her there from like
poor micro services people to people who like
born bred will die making application this way
and the argument back and forth recruit like for me were like
did they're making fun of the fact that that was so micro
right from your love you only have it's like a theme is having one object
it's only have one file each microserver should correspond to one file and that
that
and I was like no way
that's done so it is a lot that part the reason why bring it up I want people to
read about this and think about this more that's the main reason I brought
this up
good at the new area I think at the apt here
you can hear them about offers another stuff today where I try to talk about
today is what come in the cloud is now we do berta
we do horizontal scaling right that if you want to be a successful
%um were sick a nice fair give an intro
and ditches so if you want to do it there was talking about about that
compute model where I can take my app unplugging in here and plug it in there
you can do this but I would argue this is much more complicated to be able to
get that pluggable model
right this micro services
and horizontal scaling it was gonna get you there how many do you actually do
horizontal scaling in your organization today
right so how many view
you burgle scaling application today he loved asking those questions we're gonna
shut up you don't raise their hands you
all get feed in the class everybody else to raise their hand you getting a
that been I think this is what most people have to start thinking about
and my recommendation for whether its micro services are some
micro servers in white dress from for an idea like that take for a few
if you're Greenfield I would say go for this from the start
if you want to be in the pluggable compute model where you want to be able
to switch out easily cuz what you could do
this can be running on Amazon that could be running on is your
this can be running on Google depending on who gave you the best performance
right there's nothing in that model that says those all have to be on the same
clout
but you're buying latency into the whole model no matter what
writers on jatta latency your bf is building for a latent world
I agree
so you might say like okay this one thing this was an Amazon this is some
back in processing engine know that where we do our extraction and dude
and that's not in Amazon that in you're going to get better compute pricing
there
okay there go
he said a better than me
okay so many
wrap up with a min 20 and we're pleased basket I really wanted to be and I want
people thinking about like
starting to move towards that kinda service mall that was my main goal avoid
sessions
so for both people who are in the big ol after a model you like I love sessions
my
my Oracle cluster my job of course to the mic sessions really easy
sessions are the death a of horizontal scalability
move the couple your pieces we're talking about message by something else
in the middle
designed for retried in small pieces failing I think this is the thing at all
so different between a big consolidate morrow
in a big consolidating all USU everything in your big boxes always up
and if something goes down in any part a big box you like
the whole thing shit the bed I need to bring the whole I need to reboot the
whole thing
Ino horizontal scaling model its
I assume that that will go down that's why you see Netflix building like a a
funky have you heard chaos monkey
right that read they basically introduce something to rip apart their network
and their services to see how well stand up so bill with that in mind
you the architecture the web for you don't work against it
capital W web services with not Wed friendly
rest services are you the bird you the things that make you
%uh the web at and caching is your friend a surprise start without one from
the beginning
so that part that whole database discussion cash the bejesus out anything
you can cash
K when get a picture okay
moving on by chance okay let's wrap it up
if the if cloud an architect that from the beginning
I did you know you're gonna be in Amazon are you gonna be using OpenStack cloud
stack
architect wrap that way from the beginning to cam the pattern get you
more flexibility resilience incapacity or now time
AZL A's don't help when you wrap it down I wish people would
brick and get off the whole FL a thing it doesn't matter ever is really don't
matter what you want to know is how will they respond and how will they bring
things up
them guaranteeing 59 if they only give you three and they give you like
10 but back because that's how much actually cost to run your service that
means nothing
so stopping about isolation think about how you can build wrap around the
hathaway
and I'm done bank  How to Build Your Applications to Scale in the Cloud