Microsoft Azure Training - [21] Azure Virtual Machines - Part 8 - Availability Sets and Autoscaling Video Lecture | Microsoft Azure: A-Z Guide - Database Management

hi everyone this is Shauna smile with
the cloud Ranger and you are viewing the
Microsoft Azure training today is the
part 8 of Azure virtual machines and
we're going to talk about something very
important that's the vm availability
sets and auto scaling in Microsoft Azure
let's get started
the three things we're going to talk
about today is there's your availability
sets we will talk about scaling as your
VMs up and down and auto scaling which
is a scaling in and out I will explain
what scaling up and down and scaling in
and out is and the differences between
this two what I wanted to let you know
is that this is what makes cloud cloud I
mean one of the selling features or you
know important features of a cloud is
having the elasticity where you can add
and remove resources at a moment's
notice based on your requirements and as
your availability sets actually I mean
the auto scaling and scaling in general
is built on the several ability sets so
that's what we are going to talk about
today so let's get it started with as
your availability sets what we know
about as your virtual machines is that
they are similar to your on-premise
infrastructure where you have your
underlying virtualization hosts physical
hosts and you have multiple virtual
machines that sit on those physical
hosts this is no different from what you
have in your environment today
on-premises in crowd its similarly but
it's a large scale and there is a
spanning of data centers from regions
there's connected data centers
throughout the world a lot of your
organizations even on on-premise could
have a similar sort of not in such a
large scale but similar implementation
where you have a data center connected
to another data center and you have
virtualization set up in such a way over
fiber connections that if you want you
can move one VM from one location to
another for disaster recovery or your
high of Liberty scenarios and that's
what Microsoft data centers do as well
but in a much much larger scale so
having said that there are failures that
can happen anytime we are talking about
physical hardware and there are two
types of failures that can happen in
Microsoft Azure one is physical
what is not physical the physical
failure is the one where the you know
darn delaying hosts might go through
some sort of a hardware failure and your
virtual machine resides on that host and
you might have an issue with the virtual
machine the rack you are in or your
server is located in in the data center
might lose power so this is the physical
failure that can happen with your
virtual machine I mean with the hardware
and it can affect your virtual machines
and services the other thing that can
happen is failure of your virtual
machines due to hosts updates so as you
know your virtualization host on
premises has hyper-v or VMware in a 0
you have hyper-v and what happens is
that this hyper-v itself might require
patching it might require hardware
upgrade like farm wear upgrade and
things like that
and due to this updates if you have a
virtual machine on there if there's a
restart that's required then there might
be an issue with the availability of a
virtual machine so Microsoft does not do
the patching for your virtual machine I
just wanted to add that in there your
virtual machine is patched by you but
the underlying host patching is done by
Microsoft that's what I mean
now if you are in path not is
infrastructure-as-a-service the platform
as-a-service
the microsoft actually sort of patches
your virtual machine as well so what I
want to tell you over here is that to
mitigate these two issues what Microsoft
has provided is something called as an
availability set what Microsoft
basically says is do not have one
virtual machine always have at least two
virtual machines with similar role in
same tier of your application and what
will happen is that Microsoft will
ensure if one virtual machine is
affected the other one is a still up and
running so end of the day you have your
service so I'll show you a diagram in a
bit and I'll explain how that works but
some design considerations is that if
you take this two virtual machines and
put them in one hive Lib ladies set you
have to make sure they're in the same
cloud service so you have to be in the
same cloud service to have an a high
availability set to virtual put virtual
machines in there how many virtual
machines can you put in an availability
set well there's no specific limit on
that I mean depending on you know
what's your requirements are and I'm
going to go through a little bit of more
and design on the limitations but you
you could put two at least of a minimum
in a high reliability set to make sure
that you know there's some sort of a
failover and the other thing you have to
know like I mentioned is that design
wise you always want to have them these
virtual machines in the same hive
Liberty say it perform the same similar
workloads or be in the same application
tier and as usual with most think
hyper-v and I mean sorry most thing is
your you have to remember that you can
have this feature you know you can have
this feature or you can add and high
availability set while you're creating
the VM or when the VM is already created
this is like most as your stuff so let's
actually go to a diagram real quick so I
can explain this a little better so
these are three availability sets and
what I'm showing over here is a basic
three-tier application where you have a
web front-end and you have app server
which accesses database servers so what
microsoft recommends is that you have an
availability set where there is at least
similar services or similar roles are
same roles actually in the same tier
which are put in the availability sector
so the web front-ends will go in the
same availability set the apps will go
in the same availability set the
database servers would go in the same
ablility set and of course this
availability said this virtual machines
need to be in the same cloud service so
this is your wfe this is your app this
is your DB these are the cloud services
three cloud services hosting these three
availability sets where they function
with similar rules okay now if one of
them goes down for whatever reason
either a physical you know issue or a
Microsoft is doing one of the updates
Microsoft ensures the other one is left
alone as long as they're in the same
availability set and let's get into
details of that so in availability sets
the two important things both for your
exam and production you need to remember
these terms one is a fault domain and
another one is an update domain so you
know after reading a lot of material on
this all I can tell you about for
domains is that Microsoft says if you
have at least
two servers in an high-availability set
which is similar to what we had over
here at least two servers in hive
liability said Microsoft ensures that
they are in zip separate they're in
separate physical hosts in different
racks so the way this works is that as
you know in a data center you will have
racks of server right so when you put
two servers in an availability set
Microsoft is telling you not only I'm
going to make sure that these two
virtual machines are not on the same
host they're also not going to be in the
same rack so if for some reason the rack
lose power or the server has some
hardware fault we will ensure that you
are on a different rack on different
host physical hosts where where you know
if it a fixed one the other one is left
alone okay and then you have to update
domains and this one in basic simple
simple simple terms means that if you
have at least five servers in an hive
liability set Microsoft ensures that
these are updated at separate times
okay so depending on your appetite on
how much you can have a downtime you
know these are the minimums or these are
sort of the numbers magic numbers
Microsoft works with when there are
availability sets okay do you need to
put five servers maybe not maybe you
want you can get away with two because
even if you're just having two servers
they will be ensured that the update
domains are separate okay one would be
update domain 0 and the 1 we will be
updated domain 1 for example and do not
forget that this does not come for free
right I mean if you're spinning one
virtual machine for a web front-end
server and you actually have another you
know virtual machine who is running the
web front-end Microsoft is still is
going to charge you for those two
virtual machines so it's not like one
virtual machine is there and another one
Microsoft is throwing away for free it's
just that Microsoft is saying if you put
at least two virtual machines in an
every ability set I'll make sure that
there is no physical failure that will
affect your virtual machines and while
I'm updating my hosts if you have at
least two you'll ensure that one of them
is up and not all of them are
they dip at the same time okay so let's
look at this diagram over here and I
actually have six here for a specific
reason so if you have fault domain when
you create the first first virtual
machine and add it to the hive liability
set it'll be zero when you add the
second one it'll be one so fault domain
ensures that at least this two are going
to be on two separate rags and two
separate physical servers now if you
have the third one just because you can
only have two servers for the fall term
and the third one might fall into fault
domain 0 which means that VM 1 and v m3
might end up on the same rag and same
hardware so as far as fault domain is
concerned after 2 it does not make any
sense because you know Microsoft ensures
only up to two for fault domain or two
rags so even if you have 6 virtual
machines as far as the physical failure
hardware failure goes Microsoft's it
tells that hey you know I might put the
other ones on the same rack
I'm just ensuring the two of them are
separately done so if you actually had
six of them three of them would be in
the same rack maybe and the other three
would be another so three of them might
go down at the same time if there's a
power failure and the other three will
be up I'm not saying that they're going
to be in the same wrap but I mean in the
rack but their max of ensures that at
least that's how it happens and there
are possibilities that it might so you
could get away with two when it comes to
fault domains now with update domains
there are five servers so if you have
one two three four five you can see up
to VM five they're all unique means
these five servers the physical servers
that are hosting them are not going to
be updated at the same time Microsoft
ensures that up to five of your virtual
machines they'll be done separately this
is just updates this is not hardware
failure this is the update domain now
what happens when you add the sixth one
well it ensures up to the first five
with the sixth one it might go back to
zero so what will happen is that if
updates are going on these two VMs might
go down at the same time because they
might end up in the same hardware the
other ones will be done at the different
time right so that's what basically
availability sets is and what fault
domains and updates domains will do for
you and as usual you can do this from
both UI and you can do this from
PowerShell as well I'm the example over
here I'm trying to show you is the set
as your availability set this is the
command late that's used and you
couldn't give a name of this you've got
to give a name of this availability sets
and that's about it so the name for my
all these virtual machines will be
Ranger website so as long as you're
making a virtual machine and putting
them in the Ranger website I've
literally said they all will be part of
the same high availability set and as
you can look at this example by now you
can see that this is probably happening
all the a virtual machine that's already
created of course if you add this as
part of the VM config while creating the
virtual machine it will be made as well
okay so these are the things that I have
to show you now let's jump on my client
machine and I'm going to show you this
in real life and show you at what
location what is okay we are in our in
my my client computer of where I'm
connected to my subscription as you can
see what I've done in the backend is
actually create six virtual machines
their basic virtual machines with not
too much resources but I just wanted to
show you what they're doing
looking at this DNS name by now you
should be very comfortable by looking at
this what it means what range your web
all of them have the range of web it
means that they are all part of the same
cloud and let's go to the cloud service
here go to the sea ranger web
I'll show you the instances of here as
you can see that all these virtual
machines are part of this cloud service
ok now the other thing are to show you
over here is that if you want to go in
one of these virtual machines and go to
configure here you can see that I have
put them in the same availability set so
while I was creating them as I put them
in the same availability set and you can
actually see from each of these virtual
machines whatever ability said they're
in and who other
other virtual machines are already part
of it okay and the way you put a virtual
machine as part of an availability set
while creating it is this way let's go
create a new I would create a basic
virtual machine see Ranger 0:07
keep it standard a - what I'm doing over
here is that like I mentioned they all
have to be part of the same cloud
service so I already have an existing
cloud service called sea Ranger web
I'll put it over there forget the
storage again for now you can see that
there's something called here as
availability set that we have ignored
all this time what this does is that
since I created the very first one in
you know sea Ranger Ava said there is
one already over here I could create a
new one and as soon as I create a new
availability set I could give a name
over here and I'd click next next next
and this is how you create a virtual
machine in if availability set for the
first time but since I already have an
availability set all I'm going to do is
I'm gonna put it in that same
availability set click on next and click
OK and that's it that's all there is to
put a virtual machine in and
availability set and when it will be
done when you go inside and you look at
configure here you will see Syrian
Jersey row 0 7 show up over here it will
show it's part of the same availability
set the other way to see this obviously
like I mentioned is from the cloud
service since they are all part of the
same cloud service you can go to the
instances but there is something in the
cloud services where I want to show you
right now here ok
look at that update domain and default
domain so looking at this you can see
the fault domain is like two of them are
unique and that means that there are two
virtual machines which are confirmed to
be on two separate fault domains means
two separate physical hardware and in
the two separate racks with update
domains as you can see the first six
anyways 0 1 2 3 4 and 0 1 2 3 4 is total
5 so 5 of them are in separate update
domains and this one is of course
returned back to this same update domain
so
chances are when these virtual machines
Routh's are updated see Ranger observe
one and Syrian jess06 might be on the
same host and they might be updated at
the same time so that's all that there
is to have live lady set but it happens
to be one of the most important once
that you will decide on while you are
doing any production work and the reason
behind that is that it's very simple
you want to make sure that your you know
your virtual visions are highly
available and they want to be together
now some of you might have some more
technical questions that is that hey
Shawn what happens to like the on
premise clustering and you know have
lability systems between sequels and
this and all that well the high
availability over here is based on the
fact that you are going based on the
fact that you are going to have based on
the fact that you are going to have your
own mechanisms to actually deal with the
hive liability Microsoft just ensures
that you have these two virtual machines
isolated and I've ended isolated in a
sim same set to ensure the services are
there so for example if you have two web
servers going back to our example here
what Microsoft is not going to do is do
the you know load balancing for you you
have to make sure that you put them in a
node you know load balancing set you
have to make sure that your DNS is set
up properly so when your users hit this
web application or this website they are
they're okay with one of these virtual
machines going down like will the same
happen if these two database servers are
together this two sequel 2014 servers
for example you know what perhaps not
all Microsoft is saying that if you have
sequel running on BB one and sequel
running on db2 I'll make sure that these
two are not going to you know go down
all at the same time when I'm doing an
update or there's a hardware failure but
you need to make sure that you have the
logic behind to make sure that your app
servers are aware of how these guys are
load balanced or made highly available
and in sequel like you know you could do
this with log shipping you could do the
with highly bleedy like always on and
those are some things but back stuff is
not going to do that for you you still
have to do the application layer stuff
by yourself this is just high
availability for your virtual machines
it's not for your servers and you know
what this is almost similar to your
on-premises stuff you have to make sure
that you do all the application level
configurations ok so anyways that's all
that I do have for high availability set
now we're going to move into the next
thing which is scaling ok so let's talk
about scaling virtual machines up and
down and in and out and what are the
differences between them so when you say
that you're scaling up and down on a
virtual machine up means you are taking
the virtual machines for example it has
4 gigs of memory just for example when
you going scaling up you basically mean
that you go to the same virtual machine
and you go and say that hey from for my
business kind of sort of changed I'm
gonna make this now 8 gigs of memory or
16 ok so you remember this whole dynamic
memory thing and on appearances this is
sort of that thing that you're going up
and down on your resources sometimes you
need more memory so you go and add more
memory to it and the only virtual
machine this is the best part is that
you can do all this virtually it's not a
physical hardware that you have to take
things in and out and put that right so
when you add or remove resources to an
existing object like a virtual machine
you basically go or you mean that you're
going up or down this could be hard
drives this could be memory this could
be CPU this could be whatever the way in
and out works is that it rather than
going to the same virtual machine and
adding more memory you're saying that
hey I have one web front-end server I'm
not going to add more memory more RAM
more CPU but what I'm gonna do is that
rather than having one web apps
front-end server I'll have two web front
end servers or three and if when when
it's busy in the mornings what I'm going
to do is I'm going to have 10 web
servers and end of the day at night when
nobody's there
I'll come back to for example 2 or 3
so rather than adding and removing
resources to a single virtual machine
you are actually adding and removing
virtual machines itself since in Azure
it's easier to spin up virtual machines
and turns them off and turn them on
Microsoft is sort of using the best
practices where they're not going and
doing up-and-down of resources on the
same virtual machine but they're adding
more virtual machines and removing that
because this is what cloud is good at so
scaling in and out is the preferred
method then scaling up and down in as
your environment okay but nevertheless
if it's a consistent thing if it's an
application requirement all of a sudden
to have more memory on the VM it's
running on you still have the option for
up and down and as we mentioned before
that virtual machines have different
instance sizes pre-built
images for example of all these virtual
machines that you deploy things on or
you provision your virtual machines off
you can change the instance size and
let's go to you know my machine over
here and I'll show you that real quick
if I go to the virtual machine here see
ranges reserved one for example and I go
to configure you can see that my current
virtual machine size is on an a2 which
is using two cores and three front five
gigs and for some example like I
mentioned for whatever reason you say
that that's not working for me anymore
I want to bump the resources on this a
specific virtual machine you can do so
you click on save and it's it's actually
giving you the warning that your virtual
machine might be restarted and which we
totally understand so planning wise it's
important that you remember that but you
click on ok and basically you're a to
virtual machine with two cores and 3.5
gigs of memory will be bumped up to 4
cores and 7 gigs of memory and which is
pretty awesome but there's things you
have to remember while you do that that
this instant side supports that you know
these virtual machines have same number
of data discs and I've showed you that
website many many times where you can
actually go and look at all the details
from MSDN on this instance sizes and
have the number of data discs you have
to
make sure that this up and down can only
be done if they have the same number of
data discs so there might be an instance
size that will not be supported so you
can only go up and down within the
parameters of what it can do so let's go
to see Ranger 0:02 as you can see you
can go from a to and do all of this but
as your actually has way more instances
that this and there are other series of
virtual machines that you may not be
able to go to because they might not
have the same kind of data discs or you
know same numbers of data discs that are
supported so the only thing you have to
remember here for up and down is to make
sure the data discs do match so that's
planning wise you have to remember and
while we did that from the management
portal of course set as your VM size and
instance size you can specify the
instance size you want to go to so over
here in this example I'm going from
whatever instance size it was to eight
for example and this is the command let
you use it as your VM size of course you
would filter this to you know get as
your VM and the cloud service and the VM
names your capture of the VM object and
then pass this in there but this is what
you need to remember and of course you
can do this from the management portal
as well so there's not a whole lot to
talk about this but this is how you
change the instance size once you
already have a pre-existing one of
course if you have planned everything
well you may not have to do this change
of instance sizes I mean not even at all
rather than often that you'll probably
need will not need to do this at all if
everything's are planned properly okay
but that's what scaling VMs up and down
is all about but the most important one
we are concerned with it this session is
auto scaling and auto scaling with Azure
is not about you know up and down it's
about in and out which is adding more
virtual machines to a resource and
removing more and I will tell you how
this ties up to a highly in a hive
liability set so auto scaling basically
means that you are giving Ezard the
capability of automatically starting and
stopping the virtual machines
what Azure will not do for you
is as your is not going to create a
virtual machine for you you'll have for
example ten web front end servers and
you'll only be using for example two but
you'll give instructions to Azure that
hey in so-and-so timed like for example
morning nine o'clock or end of the day
of five o'clock six o'clock or 10 p.m. I
am giving you the capability of going
and turning on these virtual machines
and turning off this virtual machines
all by yourself
as URI is going to do that on your
behalf so you don't have to babysit this
based on obviously some configuration or
scaling rules that we will get into in a
bit
now scalability here is out and in or in
and out rather than up and down okay so
auto scaling is running in and out as
your will not go add more memory to your
virtual machines or change your instant
size okay
and I mentioned that many many times
because it's important all it will do is
this start your virtual machines and
turn them off so the whole up and down
and in and out needs to be absolutely
clear at this point and the VMS must be
of the same size and the same
availability set okay so if you are
using a three for your one instance you
have to make sure all those virtual
machines are going to be all a three so
it's easier for Azure to calculate
what's going on at an inserted specific
time and there is a reason for that and
how is this all controlled well they're
all part of one availability set means
also all part of one cloud service so
what Microsoft is saying that have all
your dist:10 is three wavefront ends in
one availability set and in one cloud
service and you give me some
configuration rules and I will turn them
off and on automatically for you on your
behalf
given on the parameters that you provide
which is my next point which is a
scaling is triggered based on
configuration rules and scaled rules so
you have to provide the rules for as
your to actually abide by so you can
tell them for example business hours
hey as your you know what morning
o'clock I want you to go and sort of
start all ten front end service or you
know there is some CPU utilization if
you find out that two of my virtual
machines that are providing the front
end services actually have more than 50%
of CPU utilization go ahead and spin
another five for example or all of them
so you know you can understand
automatically that the load went up so
since the load went up I want more
servers to be in the hive liability set
and you know in the load downs and for
example and make sure that my users are
not impacted and it could be matrixes
like memory usage as well queues and lot
of other configuration rules that you
can instruct as your to work on and
calculate before it goes and turns on
virtual machines on turn of virtual
machines okay and I'm going to show you
this how to do that from the portal the
thing that you have to remember with
this auto scaling is that it could take
as your up to an hour to do this what
else is your do that well as your is
telling you that hey I'm going to start
these virtual machines up and down for
you I do not want to go into a situation
where you have all of a sudden a spike
in your CPU for like about two minutes
and you know just because your CPU on
this virtual machine was spiked up to
two minutes I'm going to spin up another
ten virtual machines wanted to find out
five minutes or two minutes later that
they are all gone back to normal so as
your does not want to get into a point
where it's powering up and powering down
your virtual machines crazy because your
as you know your CPU can go up and down
depending on you know a lot of things as
you wants to have some sort of
consistency so what Azure does the
entire process to calculate to make sure
there's a consistency that yeah you are
hitting 50% of your CPU consistently for
the last fifteen minutes for example and
calculate based on that the result would
be as you're going and spinning this
whole five or six other virtual machines
this entire thing can take up to an hour
okay so this is good for macro scaling
macro scaling means that you know like I
have users coming in and there is a
consistent amount of load so I think
that I need to go and you know alleviate
some of these roles or mitigate the
whole performance issues by spinning few
more virtual machines I don't want it to
go crazy to a point that every time the
CPU is 50% for whatever reason as your
goes crazy and brings up that new
virtual machine when turns them down so
they're good for macro scaling these are
this this is good for when you have a
consistent load that is very well
measurable and this is something you
have to remember in production because
you know if you have all of a sudden a
load where Pam you have like nine
o'clock you know like thousands of users
coming in and you know within two
minutes they're going to take you down
well that's not how it will work it you
will have to give it I mean you'll have
to plan this limit more better than you
know make it go spiked up in two minutes
and you know spin ten virtual machines
or 100 virtual machines within two
minutes that does not happen so take
that an hour
I mean plan that carefully when you were
talking about auto scaling and letting
as your to it for you okay and you know
what this actually works for ninety five
percent of people this auto scaling will
be perfectly fine based on what you do
and for that five specific person two
people maybe you need to plan this out
differently or if you have the load
where you think that within two minutes
you're going to be impacted so high
maybe rather than two virtual machines
you probably should have all ten up and
running at that specific time so auto
scaling might not be for you at that
point it might be a scheduler in general
in Azure
I mean Azure has something called as
automation well you could have ten
virtual machines to start at nine
o'clock for you and just leave it that
way it's not good to do any calculations
or scaling for you it just goes and
powers them out so there are other ways
to deal with that problem but remember
remember that one one hour for the auto
scaling part okay so let's go to the
client machine and show you a little bit
about that so I'm here in my
subscription and I'm going to go to my
cloud service tab over here sorry not
over here
then I'm going to click the cloud
service and you can see there's a tab
called scale now that I come to the auto
scaling over here I just want to show
you over here that everything is
disabled and can anyone just pause that
screen right now and tell me or just
guess not tell me that why this thing is
disabled I'm assuming you have done so
I'll tell you why everything is disabled
look at this error over here before Auto
scale configuration can begin make sure
that all the virtual machines in the
availability set are of the same size
the only thing we have taken care of
this cloud service is that we do have an
availability set with seven instances
but not all the virtual machines are of
the same size and if you remember the
slides I told you that they all need to
be on the same size the instance sizes
need to be same I'll just you know as
you're just can't calculate who is doing
what so if you go to see Ranger 0:01 and
remember we actually we're demonstrating
that is scaling up and down at that time
from a two I jumped to a three all my
other virtual machines over here are a -
but my C ranges are one during the
demonstration we've jumped it to a three
and the resources changed so you
actually ended up scaling up rather than
doing anything else and the whole
scaling does not support them so I left
it that way just to show you that you
know the oldest instance sizes need to
be of the same size before you can do
outer scaling so once that's done if you
come to the cloud services you will see
under the scaling that you'll have the
scaling options and I'm not going to go
through everything but I'll show you
what will happen here you can scale
based on different matrixes and
different discs it and of course they
schedule so you can set up a schedule
and say that hey I want all these
virtual machines to be up and running in
this availability set at morning 8:30 or
8 o clock because I know by 9:00 o'clock
my employees are gonna get in or my
clients are going to get in or you know
my you know customers are going to get
on the website for example I'm giving an
example of web sites here because that
makes the most sense but there might be
other reasons it could be an app server
it could be a database server and once
again you have to make sure the hive
ability is configured by you because
Microsoft
it's not what to do that so those are
still remained away the way the
limitations are the capabilities of high
availability said that you still have to
maintain however what will happen over
here is that if you have web services
with the schedule what it will do is it
will actually make sure that at a
certain schedule the scaling is taking
into consideration and the scaling is
done for you and the other matrixes as
you can see it's enabled now so if you
click on schedule you can you can see
that you can have this scheduling
capability of scaling this okay the
other thing you can have like I talked
about is like the CPU and the queue
things like those so if you go to CPU
here you can show that hey for my
instance size my target CPU is going to
be that have let's see and these are
sliders by the way so I can say that
they have this to running my target CPU
is like from 60 to 80%
you scale up by oh I don't know scale up
by another 5 so you can set all this
information based on your business
requirements and find out that how can
you scale up so in this point I'm saying
that it's gauge ranges I'm going to have
two of them or three instances target
CPUs between 60 to 80 I'm going to make
sure that you scale up by five scale up
wait time you can actually have a wait
time before the last scaling is done so
Microsoft does give you some capability
of you know throttling the right term
will be throttling this is scaling up
and down over here but still it takes up
to an hour total for Microsoft to go and
spin up and scale you know scale it up
for you based on our slides and what we
have talked about and the scale weight
scale down time it is that hey you know
if you go and find that all the
instances between are no longer between
60 to 80 percent what do you what do you
want to do is that you want to start
scaling down you want to start turning
virtual machines off so you can say as
your that okay even if this goes below
60 percent I want you to at least wait
20 minutes just to make sure that you
know before you scaled down that you are
not doing this prematurely
go as low as you know five I would
rather have more than less so at this
point I'm doing this and with the
instance scalability the other thing you
have to remember is that when you have a
skilling like for example three of them
over here what I'm telling it that is
that with the CPU all this instance
sizes need to be between 60 to 80% for
example okay so you you it's not like
one of these instances will be between
60 to 80% and you want to fire it up
fire new virtual machines up you want to
make sure that all the three instances
that you started with actually are
between six to eighty percent before you
start on startup new ones so let's let's
go through an example so I'm going to
set my minimum over here so I'm going to
say that let's start with two virtual
machines and go all the way up to seven
okay because I'm going to use all seven
I have so basically you start with two
and seven and I'm telling that if these
two virtual machines consistently go
between I don't know 50 to 50 to 75
percent maybe yeah 75 per person we find
so what I'm saying is that if two of
this instance sizes are always having
the CPU utilization between 70 to 50 to
75% then start scaling up and when you
are scaling up is Kel I don't know two
at a time
okay so two of my virtual machines have
to have consistently 50 to 75 percent of
CPU utilization and if you find that
start two virtual machines at a time
so it will become four virtual machines
now in these four virtual machines
if you find at one point it is for all
four of them are between 50 to 75 then I
want you to scale up by another two till
you have gone all the way up to seven
maybe since I'm doing him a factor of
two maybe I should have made this 8 so
it could have gone to 2 and 2 and 2
right so but I'm just showing this as an
example and once you have scaled this up
the very first time or each time I want
you to
wait five minutes even if they're
saturated so if even if all my for
virtual machines now are between 50 to
75 percent and there is still consistent
like that I want you to wait five
minutes because who knows maybe they
will come down right so all those four
instances at that point will have to be
between 50 to 75 4 and wait 5 minutes
before they're spun up all the way to 7
instances and how are you going to scale
down well I'm going to say that scaled
down one at a time and you know when you
scaled down one at a time
I want you to wait at least 20 minutes
between scaling down and what it will do
is that it will make sure that I'm not
suddenly going down which impacts my
customers or clients I'm going down
gracefully till I'm back to the original
two over here so I'm not getting into
too many details over here and obviously
this might be a little hard to
demonstrate the CPU spikes up and down
but do play with it if this requires
some sort of a planning and let's get
into the planning a little bit the first
planning you need to make sure is that
you have those seven virtual machines
created in the same cloud service you
want to make sure all the seven virtual
machines or eight virtual machines are
part of the same availability set once
they're inevitably to say it if their
front end servers as your actually
provides you the load balancer the one
that we talked about in a previous
session so you could put them in the
load balancer and as your will load
balance them for you okay so if it was
wave front end you'd be good to go and
you know you'd come over here and plan
this out based on your requirements
either CPU or queues or memory or you
know I don't know specific time of the
day you do that as well so those are the
things you'll have to plan and the only
point where I feel that this will be a
little complicated is when it's not in
web server so you're not using that as
your slow dancer for example it's your
database server or if it's your app
server you might have to tackle that
configuration by yourself okay if it's
SharePoint app server I know that
SharePoint app servers talk to each
other automatically so as long as you
bring up a new app server it's up and
running they can talk to each other and
the load balancing is done by themselves
but that might not be same for a custom
web application that you might have
create
and they cannot scale up and down even
if you create more virtual machines okay
but that is the preferred way this is
the preferred way how Microsoft talks
about scaling or this is what elasticity
is right I mean you can you can you can
you can pull it and you know or you can
shrink it and you can make it go up and
down that this thing automatically will
be handled by you by as your that's why
it's called auto scaling okay
and Microsoft kind of prefers you do
this for at least those services you can
take care of anyways guys I think that
brings us to our end of our session
today so we talked about hive liability
set we talked about it scaling up and
down where we change instance sizes and
we talked about scaling in and out which
is ordered and that's handled for you by
Azir okay so it's important for your
production as well it's important for
your exempt purposes as well anyway so
in the first section out our training
side is as usual Club Ranger net /as
your training we are at YouTube this is
where you're most likely viewing
everything right now cloud Ranger
Network twittering is that cloud Ranger
block and of course you can always email
me at short Shawn at cloud Ranger dotnet
okay all right guys this was kind of
short than my usual sessions and I'm
trying to keep them as short as possible
let me know you know provide me some
feedback or of this works or if I'm I
know I'm trying to speak little more
rapidly than I usually do to actually
shrink them and not make them really
long one-hour sessions okay so thank you
for viewing and I will be seeing you in
our next session