Add to Watchlist

Preparing Apps for Dynamic Scaling

0 views

Citation of segment
Embed Code
Purchasing a DVD Cite video

Formal Metadata

Title Preparing Apps for Dynamic Scaling
Title of Series EuroPython 2015
Part Number 56
Number of Parts 173
Author Simkes, Roy
License CC Attribution - NonCommercial - ShareAlike 3.0 Unported:
You are free to use, adapt and copy, distribute and transmit the work or content in adapted or unchanged form for any legal and non-commercial purpose as long as the work is attributed to the author in the manner specified by the author or licensor and the work or content is shared also in adapted form only under the conditions of this license.
DOI 10.5446/20198
Publisher EuroPython
Release Date 2015
Language English
Production Place Bilbao, Euskadi, Spain

Content Metadata

Subject Area Computer Science
Abstract Roy Simkes - Preparing Apps for Dynamic Scaling Scalability is a big problem for everyone who wants to grow. In order to handle the demand, appropriate infrastructure both in terms of software and hardware should be met. What if hardware was as dynamic as a service where CPU and RAM could have been acquired when only it's needed. Is there such an environment? How can you work with it? What you should be careful of? How your applications should evolve?
Keywords EuroPython Conference
EP 2015
EuroPython 2015
Series
Annotations
Transcript
Loading...
so anyone who came to argued
that I don't think so but it's anyone uses any called service providers Amazon space just 1 to maybe what where YOU solution to and 1 uses do like containers why so if you content and you I think you have to know what was in space so you still strong our starting from all instances of but it's a just go queries and new ways you have just a new package that for 15 years of technology or something like that yeah documents it's about so some constants of about exactly but where something that has a thick content technology so take even if the Google lady was like book containers are integrated using and he was like confused about what the container technology is well learned what the packet technologies it's kind of interesting so OK let me come to that later because I have a slide which looks pretty much the there was pretty much like the communities and that's still correct so it is pretty much like that so you will notice it and we can discuss if you have any questions when we get there so any system administrators here or parallel lot of them just once I see so I can write anything I want about system administration and not that's but you certainly wouldn't was centered on right and so we can only only talking about preparing acts for dynamics can welcome and me if OK thank you for coming and 1 is wrong and I might be having this and for but it's coming back for
status and I was talking about preparing your applications for bands skills so 1st of all the outline so while making this why this idea of dynamic scaling or scaling your systems automatically is important and then what is an instance and after that we will check if your is already ready for that and I think that if you have any questions about this slide just interrupt me and ask and then we will discuss so how this scalar applications that any idea anyone has tried to scale and application what you do yeah you just put more services right you scale horizontally and you add more and more services in order to out like meet the demand so but this is more I mean what that when your when your scaling horizontally you have a server which is using like 90 per cent of the CPU and when you add a new 1 what you do is that you are decreasing the I use trick of that 1 server to 50 per cent and then you add another 1 which is you like if the person so there is a waste of 50 per cent of CPU power when that we are now using N when we increase the amount of servers the poor there you just create more wasteful resources so while we're not trying to optimize we only see that OK adding more resources the it's it's kind of like a brute force solution right it's it's there's nothing less is like yes we have a problem we cannot meet the demand what we book service and so does this really work so anyone using Amazon so if you know that you that you put more and more services and your building gets crazy right and this this all these paying for some real losses you're not really using because you are wasting and and is of course load balancer reaches standing up before your servers which you have to pay for it as well so what we wanna do is In this talk is to create an awareness that you are troubling away your money trolling away your resources it is that when you add more servers instead of adding more servers creating more computing power we believe that you have that would utilize those better remembered occurring just like what they were doing that but was like a kind of renting CPU policy for your application right so it kind of makes sense right I mean when someone is not using some kind so we have we can some other but with the communities does is that's what it's a platform as a service so you try to look at an internal model module to the Godfather you can't because its content based so what he believe that you have to utilize your servers better to make the world a better place anyone watching Silicon Valley that case so that now I look funny and then this line right so I have a youtube link so maybe you can watch it and then it might it still makes sense for you but what I have just booked at a pixel and book so what is landscape basically it's an automated and optimize way for scaling sellers vertically anyone know all about vertical scaling but so it's very similar to the let's say that you have an like a general applications and you're running it on the 1 CPU and 1 gigabytes of RAM so more users that have so what you what is the basic instinct of doing when you want to I get help search this beta so that's what that's what you do you scale but you should absolutely things so that it's not very visible enterprise manual things and other kinds of
things so when you come do it and it's very similar for all kind of website so and when you build your infrastructure you plan for either for the maximum capacity where your users are coming and other kinds of things or incremental of optimum capacity so sometimes some users will not be able to get responses but then I will say costs which I'm not able to you so you see that there are some lots of spaces where it's underutilized so if there was a total length of your emphasis will have been I mean what if you were able to plan your infrastructure dynamically based on your own so uh I remembered could this presentation this was very similar to what they have shown right the idea of better utilizing your server interest what they were doing is that as passed platform and they were just doing this for prophesy and we are doing this for servers so not just a single application that gets this performance but also you away so dynamic scaling is actually S is the way of scaling up instead of scaling up just like the other way we have talked earlier in the horizontal scaling way you add more servers in the scale of but when I don't scaling comes in you just scale of or done if let's say no more resources required your resources or your customers are living the website to go to sleep or when they come back you get so any questions so far does make sense because this like the 5th 56 minutes of this fall which is the really important upon which I should get you going so you have to better I say again you have to better utilize your services so I rate user and ready for such a change so 1st of all it does the change do you believe that this is something you are not doing already so the answer is probably you already are because you already you put your at your current applications to more power for so long and you're just you can just keep going with so the best thing about this scaling vertically is that you have you don't have to change your software infrastructure for I mean that's considered uh away when you want to put another so we want to scale horizontally what you do 1st at anyone tried but say they're providing something called a lot
of has anyone tried the 1st thing you do is to put a lot balances before a service so there is no change of going wants to return to you have to go from 1 server to at least 3 In order to be able to scale because there should be more balanced search before the service so that these laws of the distribution so when you get so you probably have all these things related to the load balancer in your application right because he is directing the traffic if you're using HDPs then probably the load balancing come from http I mean if you want to get to remove address for from the real users who will be have to using action to speaks forwarded for or something like that on or maybe you've tactical balance or something like that and let's consider what will happen if 1 user will come to 1 server and then be redirected to the other 1 so you have to find a way to share your assessment sessions like so has anyone done that market so it's kind of tricky so you can use stick sessions of other kinds of that that's fine but when someone attacks of that then it will just take on over and the other 1 will go up so it takes a lot of effort for applications to be scalable in a horizontal plane it it takes a lot of time a lot of effort and a lot of try and in vertical scaling your already ready because it's not that different than the going to a machine which has more or something so if you're using a participatory the Unicorn you can probably use scale because as I mentioned well i what Thanksgiving does is that it just library size your servers 1 moment you have 1 CPU and 1 gigabytes of RAM and the and the moment you have treaty views and 42 bytes of RAM which is a requirement for the application and when those requirements gets sold you get together and go back to the 1 CPU and the bytes of RAM and then you pay accordingly so Python PHP it works pretty well my sequel Polska there are a lot of and apart from that the important part apart from the technologies you're the other
applications where you are building your replication I also agree that we have like made a test by using WordPress which like in OK so we have like compared that our dining scaling to our Amazon's auto-scaling picture and you get like 2 times more the Trans GPS then the Amazon why is that because as I mentioned the are scaling your Service vertical and in the autoscaling what you have to start another web server and then deployed so what is going on to the other which who goes up you get lots of time for your are not same and you lose your customers because you're not able to answer them but in dynamic scaling is what's the scanning is much easier you already can so what you should be looking out for so are you like when you're starting your application by the checking the number of CPU because it's something dynamic now if you just forcing yourself to be to include yourself to out single processor then you cannot use it so are you doing and another thing is the memory so is that the constant for you like anyone remember anyone knows jealous he size but did you know that you can set the heap size that is bigger then your physical machines memory I mean if you have a laptop which is like 40 the lights of GM and for the heap size you can set 16 don't so yes if you don't increase the memory your gel application will crash but if you can it was and well of course there are some applications which doesn't work mean that anyone heard of Cassandra it crashes so here because it just preallocated memory and then when this memory of is not like there and if it's most of the times there but it cannot use it because it's things that sometimes yes Wednesday's sitting but it just tries to take the whole 16 gigabytes before needing it in it crashes so you have to make sure that not using Cassandra made so and as I mentioned if you want to really taken together all the skiing idea that you have to make sure that the tools you're using are also compatible as I mentioned at ICCV words or my sequel or but those of this tool you're using is ready for that I really do believe that even if they're not ready now they will be ready in the future because as with cloud computing walls the way that computer resources will become more and more easy to allocate into the other everyone is trying to work on these right now so yeah they will come so you should be think about those things as well because when you are let's say that you're using doctor and do you know that you can increase and you can make a library size off book con let's not say that would contains about lexicon painters you can library site you can add more CPU is more and and more this to them without that out so given if you're urinating even if you don't want to catalyst you can use it on your Trent machine by setting up a lexicon thing or the local context and you can check that if a replication is ready for it or not only you can also try possibility I mean in your application just like but we are doing it automatically but you can do it manually I mean strategy much replication so that more users are coming to your container and at some point just increase the resources see how the application will react to that to this was all my fault and just so that you might have in your mind I have prepared some questions for myself so that I can answer them easily in the book 1st let's do I really need yes I mean you don't need to do any change probably to use so why are we not using it and in horizontal scaling you have to to put a lot of effort in order to be scaled horizontally but to scalar particle well here intentionally have to do so why not use it so another 1 OK thank deliberate
OK now we have to go back to the
food it's just recap I just programmable office for that question and I give you a
and the much better way is about it crashed completely isn't perfect
and but it can recover so yes
so maybe even this is not ready for that so I ordered problem these is different than AWS automatic skills does anyone know automatic skills and so it's different because are not scaling horizontally yes scaling vertical so you know you don't need to more about this in front of your servers and you don't need to change replication in order to support distributed system about communities the related talk distance to that they're colliding in past system they're just giving you a platform where there are lots of limitations and if you're not doing your application their way you can actually use their you can naturally use their so what we are doing is that we providing an idea system so you get a ritual machines that can scale vertically automatic now the caste system where you just get a process and I guess that's the fun of important by so the question is dynamic scaling is not an alternative to horizontal scale it's complementary I mean there will come a time when I taught to CPU will not be able to handle your request or they will come a time where you would require high availability business continuity so you will have you will need to handle distributed systems at some point but you can use it dynamics can remember my 1st example when you just fallen new server he did what he out 90 per cent of the CPU was degraded to 50 per cent so if you ran that skin on both of them the moment that will be automatic scales will be the point where everyone gets to lot like 80 per cent so you better utilize your resources and end this is my fault thank you for listening and if you have any questions please ask b and I
think you know much present still in its small ball stresses instances learning about the most successful along horizontal around different it's more about stress resistance of learning more about this scalar field systems stresses and stresses the idea you paraphrase of broken off the slice of you just saw you don't you just solution of life right and so in this case so is better so creates some small small amount of those that's too big to be displayed because if you just go only 1 out of place and it's a lot of them real question you have more problems that are so small slight yet but I'm not saying the dynamic scaling is an alternative to high availability as I mentioned I mean the at some point of replication you would require high availability and what it's it's not about automatic scaling as well putting 2 summaries there there can be a tourism highly like the 1 in in those high available servers you might be using temper sample the CPU and you still get less than you pay for also but if you run dynamic scaling on them then you will get what you pay for does that make sense in my experience of understanding of it's more useful for all of the ways that might be the length of time yes it it depends I mean it makes sense for applications for me at least because most of the applications require more CPU and RAM from time to time and when the young Europe users coming more considered e-commerce site so your replication so really requires you get lots of requests and you will have to handle it some way and to do that you get by more and more so but the dining scaling yeah I had been there are some use cases and there are some use cases that it doesn't work considered that you will have you are you have a server farm where you're running in the big data computational team where all the time it's like 100 per cent of the view so yes it doesn't make sense because you are already utilizing all power you can get and that's what Thanksgiving is about better like utilizing the the resources you have I don't think so not until we can talk later if you wish if there are any way we used a learning scheme for any kind of skin into and so are you right now not done yet most focus on CPU and memory and I O is dependent on 2 things 1 of them is the uh ephemeris which which is most of friends and necessity and the some for other parts of thing is the network context we can say that blocks the which something similar increasing it's not right through the use of my question so common cycle pretty well and convinced ideas as we have more more Christian he name and it's not it's time please come with me if you have any questions which were tried to us here so been involved with anything in the
Slide rule
Service (economics)
Spacetime
Slide rule
Lecture/Conference
Query language
System administrator
Internet service provider
Content (media)
Dynamical system
Aerodynamics
Building
Context awareness
Pixel
Musical ensemble
Beta function
Structural load
Length
Scientific modelling
Insertion loss
Computer font
Maxima and minima
Mathematics
Linker (computing)
Scalar field
Central processing unit
Aerodynamics
Physical system
Enterprise architecture
Computer font
Spacetime
Channel capacity
Maxima and minima
Order (biology)
Module (mathematics)
Website
Right angle
Lastteilung
Slide rule
Server (computing)
Service (economics)
Presentation of a group
Channel capacity
Dynamical system
Scalability
Power (physics)
Moore's law
Computing platform
YouTube
Mathematical optimization
Mobile app
Slide rule
Server (computing)
Forcing (mathematics)
Content (media)
Line (geometry)
Cartesian coordinate system
Scalability
Computer animation
Software
Personal digital assistant
Mathematics
Dependent and independent variables
Mathematical optimization
Group action
Context awareness
Multiplication sign
View (database)
Mereology
Replication (computing)
Total S.A.
Web 2.0
Plane (geometry)
Mathematics
Strategy game
Central processing unit
Process (computing)
Logical constant
Structural load
Moment (mathematics)
Physicalism
Cloud computing
Database
Order (biology)
Website
Right angle
Lastteilung
Laptop
Point (geometry)
Read-only memory
Server (computing)
Service (economics)
Sequel
Distribution (mathematics)
Virtual machine
Computer
Coprocessor
Scalability
Number
Cache (computing)
Read-only memory
Lecture/Conference
Software testing
Slide rule
Server (computing)
Chemical equation
Physical law
Java applet
Memory management
Cartesian coordinate system
Particle system
Number
Word
Computer animation
Synchronization
Central processing unit
Library (computing)
Software bug
Slide rule
Server (computing)
Port scanner
Scalability
Open set
Maxima and minima
Summation
Inclusion map
Bargaining problem
Computer animation
Uniform resource name
Physical law
Aerodynamics
Office suite
Gamma function
Units of measurement
Curve fitting
Point (geometry)
Server (computing)
Scaling (geometry)
Multiplication sign
Virtual machine
Dynamical system
Distance
Replication (computing)
Scalability
Casting (performing arts)
Mathematics
Central processing unit
Aerodynamics
Analytic continuation
Computing platform
Physical system
Service (economics)
Theory of relativity
Process (computing)
Slide rule
Moment (mathematics)
Internet service provider
Coma Berenices
High availability
Limit (category theory)
Cartesian coordinate system
Scalability
Maxima and minima
Exterior algebra
Computer animation
Utility software
Order (biology)
System programming
Units of measurement
Point (geometry)
Read-only memory
Context awareness
Numbering scheme
Server (computing)
Presentation of a group
Length
Multiplication sign
Modal logic
View (database)
Mereology
Replication (computing)
Scalability
Power (physics)
Video game
Lecture/Conference
Stress (mechanics)
Scalar field
Program slicing
Central processing unit
Physical system
Block (periodic table)
Computer
Sampling (statistics)
High availability
Instance (computer science)
Cartesian coordinate system
Exterior algebra
Personal digital assistant
Computer network
Website
Right angle
Cycle (graph theory)
Loading...
Feedback

Timings

  409 ms - page object

Version

AV-Portal 3.8.2 (0bb840d79881f4e1b2f2d6f66c37060441d4bb2e)