Fast Cache, Fresh data. Can we have it all?

Video in TIB AV-Portal: Fast Cache, Fresh data. Can we have it all?

Formal Metadata

Fast Cache, Fresh data. Can we have it all?
Title of Series
CC Attribution - NonCommercial - ShareAlike 3.0 Germany:
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.
Release Date
Production Year
Production Place
Seoul, South Korea

Content Metadata

Subject Area
Area Presentation of a group Mapping State of matter Multiplication sign Cellular automaton Drop (liquid) Flow separation Expected value Type theory Casting (performing arts) Word Film editing Very-high-bit-rate digital subscriber line Personal digital assistant Speech synthesis Software testing Computer-assisted translation Nilpotente Gruppe Stability theory
Point (geometry) Computer file Multiplication sign Database Software Computer hardware Authorization Musical ensemble Procedural programming Computer-assisted translation Information security Stability theory Physical system
Game controller Service (economics) State of matter Multiplication sign Source code Mereology Computer programming Number Mathematics Casting (performing arts) Mechanism design Population density Software testing Data structure Computer-assisted translation Physical system Stability theory Form (programming) Mapping Information Inheritance (object-oriented programming) Coordinate system Counting Bit Database Cache (computing) Type theory Vector space Software Personal digital assistant Musical ensemble Quicksort
Trail Game controller Link (knot theory) Multiplication sign Numbering scheme Product (business) Number Attribute grammar Mathematics Root Envelope (mathematics) Cuboid Endliche Modelltheorie Fiber (mathematics) Mathematical optimization Physical system Structural load Cellular automaton Polygon Electronic mailing list Database Limit (category theory) Band matrix Personal digital assistant Self-organization Procedural programming Object (grammar) Table (information) Geometry
Trail Functional (mathematics) Implementation Multiplication sign Source code Survival analysis Similarity (geometry) Open set Protein Mereology Product (business) Mechanism design Casting (performing arts) Authorization Energy level Software testing Physical system Pairwise comparison Mapping Electronic mailing list Coordinate system Sound effect Instance (computer science) Cache (computing) Process (computing) Integrated development environment Personal digital assistant Musical ensemble Object (grammar)
of my presentation of a
fast test fresh data can we have both of Google has fast test they may not have press that that's and we have a very fresh states of both we might not be so fast the question is can you within reasonable budget have have little so and we had we had a cast we have a cat cat solution that we started out with some years ago I believe it's 6 or 7 years ago the at the and it had gone for a while was old stable and we thought it that everything was OK but wasn't was that maybe something correcting and how widely expectations so how was the reality from the users can we used to have a beginning maybe we had a few hundred thousand users may be 1 of 200 thousand users back 6 7 years ago and then we would have made the 1 complain a month now today we unhappy we 1 day we have less than 20 million requests so that's above 100 times as much and that maybe from having 1 complaining monthly began to have a baby a week he complaints come so we thought that maybe we should do something about it maybe maybe uh uh and slope may be that when people said it was slow uh there was something about its around so we had to set that we have to take a look at this and we set up some goes we said we would that we wanted to increase speed stability of all cancers creating an infrastructure that enables them to run as close to 12 speech on a per cent of the time during normal conditions what's keeping data 1st cut well her own the lack speed stability and what is that the I mean if you have a if everything is running so all the time he has been stability if you have and if you're in a tough 10 cases you have high speeds and then you have drops with low speeds what's called the it's than people think something is wrong and that's when we go got the complaints and that was because so it could be several thanks book but uh 1 of the reasons were of course of all cast didn't contain everything so that we would like to move on to the sinful castes but then when you don't have you had to go to the MSO and and have the MSO generate the times and that takes time and this is 1 of 200 ms per time and when you request when you request saying 16 sizes assault on hidden take you a couple of minutes before you made me do for you have have the map your you ask for the 1 on the other hand if the if everything is in the cast uh you get get it within that within the cell type in the a successful so so speed ability that's that's that's a key words so how can we angiograms um we said we would look into for 4 areas the hobbled through has a the do we have everything in stock and I mean if you have a the if you have everything in the cast you can send it out immediately so that was a goal and we had to take out and see how those all all-cash works in and how often do you get what you are looking for an when do you not just like that of the
new said nobody be infrastructure stability the this is hardware software band with database having them Christina in place some and then we would look at the end of that the excess fast access to the times fast access is about the security layer is OK of all the file how all of the authorization procedure we have a ticketing system that it worked well all we have to to deliver so to speak these
3 uh things they all of the relates to the speed of the cats the last point is
about fresh data you need to have fresh fruit on the shelves people don't want to go through the middle 1 maps and vector so we look at the year and In
this impediments and solution for all problems um and then we looked at the the access control and we saw that the the software world as we use sampled methods uh it was actually slow software we had a teaching system we had we did some counting and that and it was fairly user unfriendly to implement so we found out could better do something about that so the were a solution would be the uh to we found out that we would implement a throttling system and cut out some of the older software and controlled access through throttling the band with Raul and counting and and giving access to say that we have 3 cats and we have a professional tennis and we talk about the free uh we have we had accomplices that that allow people to take out 10 thousand times and then you shut down such a down and that that those users so we're not happy with that so by introducing a throttling and squeezing the band with a little bit and only when necessary that could that could have helped us out so we look at this overall is highly expiration of and the situation so the way we keep has press it is by living near ties expire it's and there it's a very easy way to to do that to keep the to keep the the the the cats rest but it's not it's not a very good idea is that it doesn't work well actually in in in many cases so the consequences of having a 1 week timeouts for some for the ties in all cases was the you know have you would never have full tests ties by all the time so you never reaches 100 % food test actually if you don't use a cache for week everything will expire and you have no and times and the tests that are that it's you have to ask agreement so through generates new times for and that would mean that is when you get the the MSP which is what we want the so long as they in uh the consists of presence at the time and you killed size uh because they're old I'm not because they are obsolete and but maybe what you do to partners in the question what they're talking about times this uh this is not the so you get near-miss speed rather than the high-caste that once so we have to look at that and we have to find another and you knew the solution for this Chinese together for test so we had to drop the expiration uh mechanisms the we have to we have to replace obsolete ties that means that the it should not be so that when the user the use of this should not be so that the use of this cross the times of or his program and then it goes to the and so and generally you touch we should do that ahead of time we should keep the cast of up to date all the time price times all the time the button if we do that if we do not believe ties if we had uh replays types obsolete when changes when we have changes we have maybe 10 or 20 thousand changes but they if we keep uh taking out of the year the obsolete times and replaced with new times we don't we don't have to delete ties after a given pair pair will be the expiration date that way the cash will and keep growing in keep keep growing to other 100 % and rather than the way it is today if we decrease the parents where you don't use it so much then we looked at the VMS infrastructure connections we discovered that being part of at at bigger structures molecule of the year we have and we have a castle solution that uses all normal near-miss so structure and that that was part of it so what we need to was the uh that our care services and what would that fight for resources and it would give us some instability so the solutions that was submitted to create all on a separate casts infrastructure so that and it would give us the stability wall that then it would that the fastball uh and and and stable it's so cut out that and then we introduced an we introduce a structure with only all and castles we have all maps a map source for the Constitution and we have a lot of databases critical databases to give us the data the then we look at the number of times there's sort of a number of cases that we have we have a cast but we actually had 246 ketones that's a lot of a lot smaller and and how do you get that many cases some well if you have an if you have a service 1 person was uh caspase services and for each of those we have a cast coordinate system and 1 for each formant then we ended up having 256 cases that is that's quite that it's impossible to learn uh 2 2 is impossible to do incremental uh incremental tests on on on 246 catches as we wanted to do so um when you have that many cancers or individual cases and what happens is that you get Ron having a had having a people and get data from a few uh a small number of cases and on you have people getting and the information from a vast number of cases and it gives you a low density and the a low density of fresh times so if we could have everyone go to this to a small number of cases we would get a higher density and we will not get the get the so the time axis we have discovered for quite a large number of all cases we could have done all things we could have been introduced bonded software we could could tried to do incremental updates on 246 cases which is hardly impossible so we we cut out that idea and then we said we remove them and try and reduce the number of caches and we relate and introduce analyzing many coordinated coordinate systems I think we support them almost 10 coordinate systems many of them are of the same our he's probably misuse and for all for all use you can you you can link them together you can let me say it can use um it this the 38 fights 57 for both both of those 2 E B a C a coordinate system so you can reduce the number of can no longer the number of coordinate systems uh and you can there we could reduce the number of services at least in the fast costume maybe 5 or 6 and we could do decide to offer only 1 form of we we have both J. taken the Indian uh we would cut out so did take and the ammonia PNG so we and so we end up with the uh let you are in Texas and all solutions so let's look at some of the technical details I get back to that in a 1 states that have a few
minutes OK so we looked at the access control and in the tracking system and on June organ which is the original always all the governmental and municipalities is their high priority root do throttling but only in extreme cases and for all of our models on all users we reduce struggling on the on the uh based on bandwidth but only when necessary if no problem no no squeezing of In the incremental castle operates how and this is actually a three-step procedure you don't you don't compare your database schemes team what we do is that we'll everything for every has been awake but we can do this every week it implements or take at the new changes we have about 10 or 15 thousand changes but a load them into and uh at different scheme than the 1 in that is the in production to compare and fiber changes then we can generate a list of tries to be updated and then finally updated times in the case the well I walked through steps so 1st
thing to compare the old with the new scheme but that's the 1st and then then you have uh you you only compare the same cell I use in the in this so this we have to have a vast number of tables and you reduce to and you compare only those those used mimesis and you can do all the optimization the optimization so as well like excluding near obviates like a link polygons and then you can of course look at only attributes used for classification filtering initialization within the limits of its then in step 2 you would generate the the you on identify the objects or the geometry so that's the 1st thing you do then you have define the envelope that rules that uh the the objects nearer into a uh and intersect commands and you will reduce the number of ties even further and then you create a bounding box for each time uh and you make a list of the of the ties to receive and so
the thing to do the music commands of the receiving the 1st thing you do is you truncates also those affected by uh by the object you doing the uh the seating receipts from the lowest effect level to the sooner used to generate the tightest so you and then you may have to see all the times on the highest level that is that the level and that have been truncated would be left in the that is because it would take very long time and and you cannot do that overnight for example but on the other hand you can have it you can have uh seeding process that would go to the FIL 24 7 going over time and see the missing ties and continuously when the caches and is in the high use it was showdown and during the nite when no 1 uses a cast you can speed it up and it would create the times the missing sometimes survival time from this work and all of us that well this is that this is an example of this is so the mapping authority is uh in 15 and 50 and the maps of data this just to tell you that the the comparison of the would take less than a minute Step 1 step 2 generating sub-grid and generating the list of of individual times will take about half a minute and then you do it truncates and you do the receiving and that'll the take up to 6 also in our case and reasoning of similar levels of proteins in 15 and only truncating of pseudo 16 through 19 and that's when that's that's in that case so I'm uh and we would do the same and the background seeding new commands work for us because we cannot do in take on the system to do this for that long time and for the bank uh background seeding command would still all populations so the the LS alyzing combining solution is that it takes all these the seriousness and then you do you can use some linking and you will and all that you only have to support find serious this is all new mucus infrastructure as a said we had to do throttling rather than the Roman you having a ticketing system and having a baby and of in a system for her for keeping track of users and the so we're going to of adults and introduce throttling mechanism and here you see the the infrastructure uh with the year that has only have cut out the connection to the normal test this is an infrastructure that supports only because so this is that the new situation is that you we would have still have you system with 246 in the cases we may do something about that in the future but we have a new fast test for the year and open interest that would have cut find cancers it will have to support for coordinate systems giving us only 20 individual Kessous to to do incremental update on and that will live in this case the 1st case would limited serves as focal Proquest beats 99 . 5 % of the time and there are no more than 1 the older than in the municipal this will help us getting a cast and crew keep growing and hopefully getting close to other sense making it so the whole problem with having the MSP from time to time this is from 2 or 1 of or engineers instances of Medicare supporting updating support of dating from OGI are good sources and analyzing natively we are in the process of testing these functionalities can what I just told you have been we've done this uh partly on all production system but we run and everything is truly an artist environment and within uh maybe 2 months this will be fully implemented if you have any questions you may have questions you can for your persons to come any of these people and the last 1 or you can take a look at laps but complex for them and there won't be much part of the the things get there when we get closer to implementation you find stuff about getting a fast but that's how we do this so yes you can have you can you can have it all you can have a fast catch you can press this about you will have within a limited budgets but you may have to do some offs course thank


  294 ms - page object


AV-Portal 3.19.2 (70adb5fbc8bbcafb435210ef7d62ffee973cf172)