| 00:00 | Uh huh. Mhm. Okay. try to ask. Mhm. |
|
| 00:47 | Uh huh. Okay. Men itself questions on this 1? Yes. |
|
| 01:04 | yeah so we got a problem. the second problem you have the fights |
|
| 01:10 | the question. Uh huh. The are a platform on windows. You |
|
| 01:21 | easily do that by application for a city. There should be a I |
|
| 01:32 | to the doctor uh come back with . They're not actually going to the |
|
| 01:44 | . You should you should have your military there on the customer. Everyone |
|
| 01:49 | the second from that juicy. There not be any content about it. |
|
| 01:54 | she creates separate photos from its So I don't get it. Uh |
|
| 02:00 | we don't have access here on the . So the way it works is |
|
| 02:06 | you guys use the cluster earlier in dust and the assignment. Uh |
|
| 02:12 | But once you are done I don't for first assignment they need to submit |
|
| 02:17 | goals because it's just match monday I think you're giving the codes. |
|
| 02:23 | I think in the in the future consignments you need to modify those back |
|
| 02:29 | those ones we need to be mortified hundreds of years before this one you |
|
| 02:38 | need to submit screenshots will be uh this long. Oh so there are |
|
| 02:50 | was wondering if you guys found a hydro class on Tuesday last week |
|
| 03:09 | Yeah. Yeah so when you have modifier coats. And so yeah I |
|
| 03:18 | no pleasure or running your calls to that the results you report actually from |
|
| 03:25 | code and they can reproduce. So why we wanted to submit the calls |
|
| 03:32 | country and comment If you see obvious and gesture. And the other thing |
|
| 03:45 | think most, but I'm not sure has yet accounts. So can one |
|
| 03:52 | you here through our on the net not yet creating their own accounts. |
|
| 04:01 | flu shots so you can get But I think most of us so |
|
| 04:10 | you all have accounts. Okay. good. All right. So uh |
|
| 04:17 | will start to talk about what's relevant guess for the first assignment. So |
|
| 04:27 | supposed to from some clothes and it's about getting your familiar essentially first would |
|
| 04:34 | initiative environment and using the storm They have to some jobs accused and |
|
| 04:40 | what happens. And the other thing if you look at the assignment, |
|
| 04:48 | starting up in some ways from the and just trying to understand the behavior |
|
| 04:54 | the single friends or not being very is just focused because in general You |
|
| 05:01 | make up for four single core or processor performance by having a lot of |
|
| 05:07 | really just started first to use a thing well and then you go from |
|
| 05:14 | get more resources. So that's kind thinking behind you starting from and for |
|
| 05:24 | little complicated as possible just using the core read and try to see how |
|
| 05:30 | works. And then we go from hands. Thanks more complicated unfortunate. |
|
| 05:38 | the question I have like God for , Oh, that's all I don't |
|
| 05:47 | any particular books because the White out this class as usual. Maybe, |
|
| 05:55 | know, one chapter from one one chapter from another. So transpired |
|
| 06:01 | important to them give you the slides give you the recording. So we |
|
| 06:05 | talking to the sides and not just to figure out what was the fight |
|
| 06:10 | time. And most of the I hope conception it is not so |
|
| 06:17 | to remember and we don't do things are incredibly detained, I would say |
|
| 06:24 | my view, but maybe it is you. So, so josh comments |
|
| 06:29 | students and he knows how to speak for. Uh, but let's try |
|
| 06:36 | keep things that try to make intuitive you what's going on on the subway |
|
| 06:42 | of the kind of uh huh advocating uh maybe pushing you to try to |
|
| 06:51 | about what happens and tried to explain then we get the people out there |
|
| 06:55 | seem to make sense to us. in my house, not the right |
|
| 07:01 | models from what month. So or um, there was a few slides |
|
| 07:11 | Snapchat in seven of them do not to and I will talk to a |
|
| 07:17 | of them to know that. So a little bit about three cold serious |
|
| 07:23 | to run. I remember writing the . What Brett strumming these coats. |
|
| 07:31 | before then I'll talk about what I talk about last time and I'll come |
|
| 07:38 | and talk about this more detail but best three difficult for different purposes that |
|
| 07:45 | talk about developing and then sometime or of the functions together. So so |
|
| 07:55 | when you try to assess both good code is and a good new computer |
|
| 08:01 | is a number of different things that to focus on. Sometimes if you |
|
| 08:07 | aggregate in all this is an application has lots of stuff in. It's |
|
| 08:11 | idea, the thing that is relevant you. Um but many times you |
|
| 08:16 | try to understand the various ports on computer system and how it before. |
|
| 08:24 | The two parts that uh, you , the basic ones generally is so |
|
| 08:30 | the core and how the functional units . Um, and then also the |
|
| 08:36 | that's the code actually make efficient instagram that. And the other part that |
|
| 08:42 | the weakest part, certainly on a system laptop service but not is a |
|
| 08:50 | system. Yeah. Even with the connection networks and that conflict. So |
|
| 08:58 | just looked at the single thing, way the system is really the american |
|
| 09:04 | , but it's also good to understand they accept course for the process itself |
|
| 09:12 | . So the first benchmark, I'll about them a little bit is thing |
|
| 09:17 | shows up the number of times because so simple and making complication and trying |
|
| 09:21 | understand that. And the other two to be with other memories is report |
|
| 09:28 | I'll talk to them and a bit in detail. But first thing |
|
| 09:33 | So timing uh, seems to be trigger thing called the timer. And |
|
| 09:37 | know, in the start the program start section of the code and call |
|
| 09:42 | after this section is done. That's . That's the time it took. |
|
| 09:49 | it is unfortunately not quite that simple the various timers either the sort of |
|
| 10:01 | included all kinds of things. This the time of day kind of |
|
| 10:06 | Whatever they capri is when you start and whatever the clocks has done and |
|
| 10:11 | kinds of things that happened between these times down. So it's a really |
|
| 10:16 | too engineering code to perform well. not very helpful because really need to |
|
| 10:25 | at aware distance perhaps what are the . So then you need more detailed |
|
| 10:31 | on what actually happened. And for there is also a number of functions |
|
| 10:36 | are not so easy to figure out they actually. So here is kind |
|
| 10:43 | two of the kind of potential function that gets you some more detailed information |
|
| 10:53 | exclude things that affect the wall clock . So what other jobs are running |
|
| 10:59 | the same city at the same time even other processes for programs. It's |
|
| 11:03 | the one who are interested in, affects a challenging time kind of |
|
| 11:09 | So it? S delicious, monitors your code does and that's true for |
|
| 11:14 | timers quality and they get time type connection now. So all these timing |
|
| 11:29 | , they are essentially driven one way another and depends on the computer |
|
| 11:36 | how it's driven from crystal and computer is the basic source, you |
|
| 11:46 | the processor and various parts of the run for different clocks. Not the |
|
| 11:52 | thought by having caught multipliers and dividers whatnot. Whatever is useful for that |
|
| 11:58 | of the software. So there are different talks but essentially that's kind of |
|
| 12:03 | source of time unlike the phone and know, check with the network and |
|
| 12:07 | get some global time from some reference and yes, if the curious and |
|
| 12:14 | network could potentially do that. But think that these times even if it's |
|
| 12:19 | sitting by itself and it's based on . Now, the other thing is |
|
| 12:27 | course when they want to, you , start a timer or oclock |
|
| 12:34 | it basically is intrusive. So if want a high resolution, it affects |
|
| 12:43 | running time of the courts have tried the time. So there's kind of |
|
| 12:47 | trade off on the resolution on the that use and mm the inclusion on |
|
| 12:58 | so most of this then have a of setting. How often you kind |
|
| 13:07 | read the car. So those and you look at the indian express and |
|
| 13:15 | Russia has done a few times over years it's trying to figure out what |
|
| 13:20 | come of the deformed from each one those function calls and how do you |
|
| 13:25 | control But generally do not even And clocks that's on the sporty. |
|
| 13:35 | can get the notion of maturity instructions context but it isn't really coptics so |
|
| 13:42 | can be fooled by the furbish that being used to describe these functions. |
|
| 13:48 | it's something that again that gets Uh huh. After a given interval |
|
| 13:54 | is either predefined or you have somewhere you can define something. I want |
|
| 13:58 | have a middle section. I want have a second and it counts how |
|
| 14:02 | times it's been updated without in kabul then the earth. So it reports |
|
| 14:07 | the number of times that function call been variable but corresponding to the function |
|
| 14:13 | has been updated. And then you it to time the design into want |
|
| 14:17 | set it up after this. That I think that's what dyslexia is. |
|
| 14:27 | water. Yeah. Oh yeah no . So here is just the start |
|
| 14:37 | one of this clock. It's called . And that has the predefined but |
|
| 14:44 | can also set everybody's variable. This but again and then it means that |
|
| 14:49 | corresponding variable or counter it's updated with frequency. So then tells you how |
|
| 14:55 | updates there were and then when you the period and figure out what the |
|
| 14:59 | was. Yeah. Um so here the one thing to do, but |
|
| 15:06 | want about and I used to be that some issue define your own variables |
|
| 15:11 | keep stuffing that you need to have bits in the variable because in front |
|
| 15:16 | the critical corporate sauce, so this of maybe on this side are but |
|
| 15:24 | , I know this is a so but this basic and that's not |
|
| 15:27 | mistake, that is often done when asked at the time and coat you |
|
| 15:33 | and that's fine. You know, want something that doesn't run forever, |
|
| 15:39 | if it's made responsible again, you , Maybe 100 500 matrix or something |
|
| 15:46 | smaller. And so I kind of members, you know that and then |
|
| 15:52 | the operation, So it's 100, basically hundreds of the third power. |
|
| 15:59 | if there's not familiar and these processes are during millions of instructions per |
|
| 16:08 | So if that code runs well, know, it may be is finished |
|
| 16:14 | you get an update on your function at the clock, so your best |
|
| 16:20 | , but so you need to be of, you know, the workloads |
|
| 16:26 | trying to time and ensure that the of the updates are sufficiently small, |
|
| 16:35 | at least you have probably at least updates for so on the class before |
|
| 16:41 | job is finished. Otherwise would be inaccurate. So that's kind of a |
|
| 16:47 | mistake takes a small problem and there have actually done before the clock gets |
|
| 16:55 | . Yeah. Um, so other things is now the in terms |
|
| 17:09 | time you get the resolution according to if you want things up till every |
|
| 17:17 | . Yes, that is the You get the report. Now trying |
|
| 17:22 | figure out ultimately it may be of and to figure out the efficiency, |
|
| 17:29 | many psychos did it take? Because you know what can be done in |
|
| 17:33 | single instruction and you know, I to work to be done, you |
|
| 17:37 | figure out how many cycles should Of course nothing is 100% efficient. |
|
| 17:44 | at least that's the starting point. converting it this time that the primaries |
|
| 17:50 | two cycles is not entirely trivial because most systems today the cop is not |
|
| 17:56 | unless you're forced to be thanks. because the clock rate is controlled by |
|
| 18:06 | that senses the temperature on the various of silicon and if it seems to |
|
| 18:12 | a little bit getting about to be hot when they're trying to predict a |
|
| 18:17 | bit ahead of time, then the will stop. You have no |
|
| 18:22 | Normally that's you're forced to call the of things will not go on. |
|
| 18:27 | that's one said things are variable and also have to make sure you know |
|
| 18:31 | her particles in the otherwise. So may decide that there's enough headroom in |
|
| 18:36 | of cooling and temperature and I think up the clock so it does it |
|
| 18:40 | a period of time and then slows down and that happens during the execution |
|
| 18:44 | the job. So clocks are not kids for the duration of so that's |
|
| 18:51 | of things to be well and I this is the other good time and |
|
| 18:56 | think that so you guys were used get time showing a little bit |
|
| 19:03 | So I think that was about how deal with timing of course and |
|
| 19:12 | depending on what to do in class one thing but domains related life. |
|
| 19:17 | also want to use that kind of about how to use this part |
|
| 19:21 | You can get close to what you to. So now on to trying |
|
| 19:27 | understand both are useful. The processor for whatever he wants to run or |
|
| 19:34 | well they're cold. Does this is have to measure and really johan needs |
|
| 19:40 | have the system do something in order measure sitting there either. Probably doesn't |
|
| 19:45 | you much if I want to So that this matrix multiply is one |
|
| 19:50 | is very afraid contributes to figure out good how much or what fraction of |
|
| 20:00 | to be honest. That's what people . You know how much you can |
|
| 20:04 | at suspect for a processors as the . Credit for getting incarcerated. |
|
| 20:10 | It has what's typically say No one fused multiply hair arithmetic units. That |
|
| 20:18 | it doesn't have and multiplying the single as a minimum and then they have |
|
| 20:25 | sort of factor instruction sets or seeing instruction set. So that means in |
|
| 20:29 | single market act it was a bunch these few months. So that's the |
|
| 20:34 | you can figure out what is the performance of this step. And then |
|
| 20:39 | take your cold. In this case multiplying you run it and then try |
|
| 20:43 | figure out how far off or you he treats her cold and figure out |
|
| 20:49 | close can you get to this peak . Yeah. So no demand responsive |
|
| 20:57 | again is something. So yes functional and how well our close you can |
|
| 21:05 | and the other two that is kind uh red box around are Thanks for |
|
| 21:10 | memory system and I'll talk about those codes that you were used in their |
|
| 21:17 | . Then there's the industry they tend use so called spec tank marks. |
|
| 21:22 | those are Uh huh complex but they physical brands are typical usage of the |
|
| 21:34 | . They're trying to sell some. to give you a representative number for |
|
| 21:38 | the market is for that particular We don't do some because they are |
|
| 21:45 | free. That's the one thing. it costs money but for these purposes |
|
| 21:51 | perfectly finest right here help you understand core performance and the memory system performance |
|
| 21:59 | decided perfectly fine for that purpose and think that and then you can do |
|
| 22:06 | society more complete application performance step. they're sort of higher level things in |
|
| 22:14 | this is called and asked french part once and for those tend to have |
|
| 22:20 | some of the equation systems are distorting something more perhaps useful. Then just |
|
| 22:26 | very core pieces like might provide a of matrices. So here's kind of |
|
| 22:35 | of the so called D jam for position, you know, maybe smart |
|
| 22:41 | and this is a very standard So the public domain or open source |
|
| 22:46 | of it. Most computer vendors like I do have an AMG the ones |
|
| 22:53 | the processors, they make their own optimized for their own processes. So |
|
| 22:58 | is a very common function you will in both commercial libraries as well as |
|
| 23:03 | source libraries and this is kind of general formulation what it does um the |
|
| 23:12 | thing, necessary performance and other comments . That this also useful but because |
|
| 23:19 | optimized code can get very close to theoretical peak performance that's produced pretty much |
|
| 23:26 | cycle that is available. It also to be to dissipate the heat. |
|
| 23:35 | it's also stress test persistence to figure if with the cooling system and this |
|
| 23:42 | system also and uh that's a little to the side storage. So once |
|
| 23:50 | a time years ago I was involved bulletin Larger cluster at the time was |
|
| 23:56 | votes. And you specify because the was significant so you and you need |
|
| 24:05 | make sure that you have enough power the computer and uh, mm then |
|
| 24:14 | where this benchmark and it's slightly different and it's under our stuff. We've |
|
| 24:22 | the campus circuit breakfast for the whole trip. So, and because engineers |
|
| 24:30 | quite figured out, they told us , you know, that these fire |
|
| 24:34 | was supposed to be And that was 50%. So if it's for |
|
| 24:42 | you plug it into a world Nothing happens presented it personal megawatts that |
|
| 24:47 | don't necessarily have enough professor. So a good test for two things. |
|
| 24:53 | out how well you can use the unit. But also stressed this for |
|
| 24:59 | car. And this I think is the same much, I think. |
|
| 25:04 | don't know if you plan to say about it and they're more or |
|
| 25:08 | Okay, so here's um, just , old table by now, but |
|
| 25:15 | one of the two greenish Collins going the Dijon and there's such efficiencies that's |
|
| 25:21 | fraction of peak on whatever systems less the left column. So if you |
|
| 25:27 | at the numbers, they're typically And some of them are even |
|
| 25:34 | So how can you get more than mistakes. No, so the way |
|
| 25:46 | usually figures their sufficiency, thanks a cartridges and the nominal clock frequencies below |
|
| 25:57 | for the term of trump is if happens usually insist when they figure out |
|
| 26:02 | the maximum is used in armor But then the system is itself actually |
|
| 26:07 | at the higher top frequency than assumes votes. So that's why they get |
|
| 26:11 | performance, 50 accidents. That's why end up with sometimes these folk numbers |
|
| 26:17 | looks like use the base as well the right face. The other thing |
|
| 26:27 | another green columns such as people in States stands for performance and impact and |
|
| 26:35 | come back to that later in the . And another side today too. |
|
| 26:39 | what that is is based solely Gaussian if you remember you're solving systems of |
|
| 26:46 | . So it turns out to be of the benchmarks has been used for |
|
| 26:52 | to rank computer systems because they just to play after all turns out to |
|
| 26:56 | useful in so many applications of people . That's the risk and the point |
|
| 27:03 | this and they will come back to later. Not too many problems but |
|
| 27:08 | the way you do guys in the core routine and getting it done |
|
| 27:13 | job done is make it hard to . So that's and you can see |
|
| 27:18 | high performance, low impact numbers are or less similar to the responsibility. |
|
| 27:23 | bit of simple specific but predominant. makes responsible. Okay. And I |
|
| 27:32 | about and this multiplication and the Mhm So um the next thing that |
|
| 27:42 | so called stream benchmarks has been around years and it's intended to test the |
|
| 27:49 | system in particular the main so and if you very simple operation, I |
|
| 28:03 | I should comment and that's what the are. That's the showing that processes |
|
| 28:10 | got contrast invested but memories in fact get much faster increase has been increasing |
|
| 28:18 | for many, many years and that's testing the memory system is important because |
|
| 28:24 | turns out that I would say most are LTD violent capabilities of the memory |
|
| 28:33 | . Coming back to the matrix multiply , sort of matrix multiply. They |
|
| 28:37 | of agreed that number of operations is of on those columns. So thank |
|
| 28:44 | . But the matrices are in square . Right? So on average |
|
| 28:50 | And operations for Ellen Yeah, that's decision can benefit from fascism. Memory |
|
| 28:57 | of because you have used data. applications don't and then they become limited |
|
| 29:04 | the nervous system. Mhm So here what extreme thing is it's festival. |
|
| 29:11 | simplest thing is coffee. Just made from memory. Right. And it |
|
| 29:17 | be a little bit more in this of variables if you're multiplying and the |
|
| 29:24 | one is doing an advance of some then the trial trial and then something |
|
| 29:30 | was used or still is used in application causes one of the core instruction |
|
| 29:37 | dealing with the right so this is a very simple process it seems, |
|
| 29:44 | it comes up to get performance. started not I don't think so. |
|
| 29:50 | come? It's not simple. Well huh Part of the reason there is |
|
| 29:57 | data review so you cannot get any from fascists. In fact cashes tends |
|
| 30:05 | get in the white and make things . So that so pretty much all |
|
| 30:12 | out there has ways to bypass caches you have operations of this type. |
|
| 30:23 | that's one um Oh now the other is that memory main memory since the |
|
| 30:32 | was supposed to testify member system. main memory system is very far away |
|
| 30:40 | from the functional units in the court maybe hundreds or even thousands of cycles |
|
| 30:46 | way. So if you discover that don't have the data and so read |
|
| 30:53 | story is a little odd. Daniel have to wait until the cycles before |
|
| 30:57 | get the data. So that's the part shows stream is very late and |
|
| 31:04 | sensitive so that's right. Yes, a both pre fetching doesn't work too |
|
| 31:09 | Then they get caught five this statement you have to wait because it's not |
|
| 31:14 | fetching works well may still be in cash maybe but it doesn't oh come |
|
| 31:22 | and the other books have on the I guess is something that definitely maybe |
|
| 31:28 | super right advocate policy if you have fashion system. So memory, you |
|
| 31:36 | what I recently but that's just have called cash funds and this is something |
|
| 31:41 | everybody. This for instance collection of item or other systems luxury business, |
|
| 31:48 | atomic unit when leading remember system. I may be 64 values or something |
|
| 31:56 | upon But it may be a two 64 and 1285 or something that chunk |
|
| 32:02 | read. And the way the caches something is that you can't just Write |
|
| 32:09 | memory say one word of the cash , you have to write the whole |
|
| 32:14 | . So in order to do that you want to update the fashion in |
|
| 32:19 | that had to read the whole cash from memory then updated directed back. |
|
| 32:23 | when you want to just to write you're actually in for the read to |
|
| 32:27 | the thing and frank is back. in that case that's what you pay |
|
| 32:33 | extra penalty penalty in terms there's a operation didn't want to happen. So |
|
| 32:41 | number of things that makes it difficult get very good performance of something that |
|
| 32:46 | input a stranger. Uh, it's old data but it's not still a |
|
| 32:54 | devastation of that. Even computer companies have large staff knows that this is |
|
| 33:00 | important type of operation and they want get it to work well but so |
|
| 33:07 | you know, two generations back new , unfortunately. But as they start |
|
| 33:13 | empire seven that is kind of the of the line process and not just |
|
| 33:17 | medium from anybody. A very important is very expensive. But for this |
|
| 33:24 | benchmark, the basement and execute was using 30% of the memory that after |
|
| 33:30 | the optimizations. So let's just point that it's hard. It's not just |
|
| 33:37 | for any one of us. It's also in a good education. |
|
| 33:43 | And then I had another side from of these things we played around when |
|
| 33:47 | main point is uh in terms of Steam, you make sure that you |
|
| 33:52 | on the race are large enough that are actually in maine memories, you |
|
| 33:58 | around and do things bunch of frantic statistics. And if you need to |
|
| 34:03 | sure that the race are large So after the first look, |
|
| 34:09 | everything is not in cash. It has to be a member. So |
|
| 34:14 | why the typical says that you want raise that about after memory size just |
|
| 34:19 | be sure. But there's Representative, ? That's just an eight member |
|
| 34:26 | And so you're just a little example small processes. You know, it |
|
| 34:32 | be very small enough that the or the first wave of cash. So |
|
| 34:36 | you run things, you basically see long past before so you can use |
|
| 34:40 | for that good. It's the next after I figure out what it is |
|
| 34:45 | give it what is this trip and on. This is future And this |
|
| 34:57 | I guess just the grandfather suggest Okay, Now after then there was |
|
| 35:06 | other one that's not us comes and stands for he got up this process |
|
| 35:16 | stream uh huh very gentle benchmark for nose member assistance are designed to do |
|
| 35:29 | well, Appeative speaking as well as if you access data in successive memory |
|
| 35:40 | . So there's an increment for That's one thing should go up. |
|
| 35:47 | not typical in most applications. So you have an agenda, certain soaring |
|
| 35:54 | problems or parts problems you and no well defined memory access planning. So |
|
| 36:03 | that reason I'm use call that's benchmark is randomized access, read, write |
|
| 36:12 | memory, there's a totally different access so on. Is kind of showing |
|
| 36:18 | memory system at best and gaps in sense shows more of a typical member |
|
| 36:26 | performance. So and then I have human size here's something you'll never get |
|
| 36:34 | same thing. Again, we need make sure that the data array that |
|
| 36:39 | working on the start enough that this in memory and things always goes to |
|
| 36:43 | memory. Now to flash access and it's a little bit just hardly do |
|
| 36:48 | reorganization in terms of accessing it. don't get to do it for the |
|
| 36:52 | interested If you go to the link the client describing them in detail or |
|
| 36:59 | can take a look at power. could do the random organization of |
|
| 37:04 | Perfect. Yeah. And there's so you pretty much sent already. Just |
|
| 37:14 | large enough. So it's basically randomly works to memory a very long |
|
| 37:23 | . Okay. We also show that , I had one what I want |
|
| 37:28 | point out so what to, we'll out when you do this to. |
|
| 37:33 | more. Thanks. So the stream codes, it's nice enough that tells |
|
| 37:42 | the access rate to memory in terms gigabytes per second. And remember that |
|
| 37:49 | suggestions last time. Okay. A insensible kind of memory. It is |
|
| 37:54 | the user using its memory channels. is. So from that you can |
|
| 37:59 | out with the pitch memory boundaries. typically I think on 19 2 is |
|
| 38:08 | about 100 megabytes per second. But it's in that order. So |
|
| 38:15 | with and tell you, you know , I'm in a big eyes per |
|
| 38:18 | to get for running these benchmarks and you can see your little traction or |
|
| 38:23 | that you got the gaps. Doesn't do that. The number of updates |
|
| 38:31 | second. And then you need to it bandwidths given the data type and |
|
| 38:36 | by cities for access. You have do a little bit of that in |
|
| 38:39 | to make the numbers comparable. Yeah. So this side but then |
|
| 38:46 | not strictly stream and guts but this from this top times under this based |
|
| 38:56 | it's the only guy in this Some regulation on possibly enormous dense |
|
| 39:01 | the best of matrices that all of numbers and the S P C B |
|
| 39:06 | G is an intuitive solver for what's difficult to use for what bars making |
|
| 39:17 | . So what I wanted to get focus on Is based on the percent |
|
| 39:22 | tea in the % Tea 2 Calling for Other verses the S P C G |
|
| 39:29 | so you can see this. They co they got 75, 85% of |
|
| 39:36 | . Again. It's like nature's not because that's court. What's being done |
|
| 39:42 | look at a specific comment That far and it's 123% of people. So |
|
| 39:50 | tells you it's essentially and the memory is being supported something the right thing |
|
| 39:56 | kind of the guts comin and there a total estimated public or syncope. |
|
| 40:05 | that's why it's important. And that's of the things I wanted. I |
|
| 40:09 | you to do both of these benchmarks be in the sea Important. The |
|
| 40:14 | access patent is it's not that So factor 30 factor 100. |
|
| 40:25 | Mm. Okay. So that was the three benchmarks and the questions on |
|
| 40:36 | of this strength catholic. So mhm has it. Believe this. |
|
| 40:48 | 64 access to state of But it how the precious works of these are |
|
| 41:01 | cash but it might be that you get all the data that the fashion |
|
| 41:08 | not just imagining the question so an that's why we have sizes for the |
|
| 41:14 | list you can play out and the over and just see how much more |
|
| 41:22 | . Yeah it is likely that once pass you cannot think that size. |
|
| 41:26 | smart was not enough most of the . Yes. Uh huh. Uh |
|
| 41:37 | . Right. So this one it never visit the former just very much |
|
| 41:41 | enough actually one and just be great in Rwanda or Some messages that have |
|
| 41:54 | listen to that performance doesn't do that And then the door one. This |
|
| 42:01 | a fact. That's okay the Children is a fact. So and it's |
|
| 42:12 | same that's and a little bit considering question that's right element the various of |
|
| 42:25 | . Yeah yeah yeah. Okay so interesting for your own said the other |
|
| 42:36 | benchmarks to the test is caches and on that screen is the most common |
|
| 42:41 | it's a simple one. And that my perspective is most real applications are |
|
| 42:49 | So small that the 15 caches so main memory performance is really what's relevant |
|
| 42:54 | most applications then you can tweet things this and try to understand the architecture |
|
| 42:59 | then you want to play out smaller You can use strength for that. |
|
| 43:10 | . The other question. So then next thing I want to talk about |
|
| 43:16 | shit simple models that try to characterize performance and systems so as known as |
|
| 43:27 | rope line model. I was the or a model of things that start |
|
| 43:32 | people came up with a bunch of ago uh basically tried to capture both |
|
| 43:41 | system performance and when is safety you versus member system limited. Uh and |
|
| 43:49 | to also then characterize applications. What's is it likely to be emitted by |
|
| 43:55 | member system or by the CPU or court performance? So that's what's called |
|
| 44:04 | flying that is you know islam tip and then there's the roof that the |
|
| 44:11 | line is kind of things from the system and it was on the line |
|
| 44:17 | the possible performance. Of course the . Yes. Thanks. And what |
|
| 44:25 | on the horizontal exercise I think that with after how many operations with a |
|
| 44:32 | of so seeing this kind of Uh huh. So here is this |
|
| 44:47 | of a contemporary as I said that to starting up trying to figure out |
|
| 44:53 | better reuse does the cold. You potentially have questions for you to write |
|
| 44:59 | code operated. Take advantage about because because make this multiple instances when average |
|
| 45:07 | operation that they like doesn't mean that pays in cash. So you can |
|
| 45:12 | from a hurricane if they have a you still have to go to memory |
|
| 45:17 | technology. But this sounds in sort inherent in whatever the code is is |
|
| 45:24 | attached. So you're a bunch of . Person don't even the best whether |
|
| 45:33 | very well are you from? I'm arrested. Uh huh. And there |
|
| 45:46 | . Yeah. So that's a useful to try to set expectations for what |
|
| 45:54 | conceive by the department of systems. now and talk through this model and |
|
| 46:00 | bit and on how the different system attacked trying to what you can |
|
| 46:07 | This kind of picturing of thank Probably tell the system relative to expectation |
|
| 46:14 | and other typical facing measured in these . So this is a simple tenure |
|
| 46:23 | the theoretical peak performance Tpp sometimes to that you just do the way I |
|
| 46:31 | a little while they go inside and at the heart frequency. You can |
|
| 46:35 | out how many operations cough it can . I mean of course you're not |
|
| 46:42 | possible number that's the system. Um I guess this is what I already |
|
| 46:51 | . So and typically the first stop to buy something. Everybody has that |
|
| 46:58 | it down. So it's to do same bed front and established. |
|
| 47:07 | And about that. So what so I mentioned that before that there was |
|
| 47:17 | system today that you use and they refuse multiply add instructions. So they |
|
| 47:24 | multiplying at a single construction or too carrots on them as you have a |
|
| 47:31 | but each functional unit. Ege it's kind of mask quite terrible physical new |
|
| 47:38 | and that works great. It makes want to find an even number of |
|
| 47:42 | a bunch of places. Perfect. not everything is like that. So |
|
| 47:48 | it has an unbalanced non balance between a multiplier, you're never going to |
|
| 47:53 | that performance. So that's what the line here now. Uh Yeah. |
|
| 48:00 | huh. You don't have an equal of as a multiples. You can |
|
| 48:04 | get the peak. So you have stay down a little bit from what |
|
| 48:08 | possible is for that and then the one here. So that and many |
|
| 48:17 | system today to have these factors in instructions that a bunch of them. |
|
| 48:22 | that factors into the guarantee. But you cannot use those um the possible |
|
| 48:31 | also seeing the Earth's core, there no reliable because this feed in terms |
|
| 48:39 | nothing worried about the memory system of how much things about the function. |
|
| 48:46 | today there was many of the modern have What's known to say the x |
|
| 48:55 | similar name. five throws. So can do 16 Operations were 32 operations |
|
| 49:02 | a single destruction. So if you're doesn't characterize then I'll get one. |
|
| 49:10 | you get 1 16 Less than 10% the performance. It's a dewdrop relative |
|
| 49:16 | the peak if you can use seeing instructions of such as Oh, what's |
|
| 49:23 | ? The next thing is what some level colonels and that is that if |
|
| 49:30 | can't concurrently at the same time president slow the load, store operations and |
|
| 49:38 | operations, then you have to wait those that sort of thing. Place |
|
| 49:43 | you use the function. So you have this instruction level parallelism performance |
|
| 49:49 | Again, that's a good problem for . So the very big performance that |
|
| 49:58 | start to look at and understand your , the real peak number may be |
|
| 50:04 | lower based on the code. And part of the thing that I want |
|
| 50:12 | to be become aware of and learn this course to figure out when they |
|
| 50:17 | at the cause I just got this and I think about it, you |
|
| 50:22 | , what could have of these various could have defected to perform. Mhm |
|
| 50:29 | , no, we can also So hope here and as I mentioned talked |
|
| 50:35 | struggling if briefing thing doesn't work out way. So the foundations LTD so |
|
| 50:42 | produces That performance and next one is . So it's a little bit more |
|
| 50:51 | . Um Mhm This new my familiar anyone or all and not no assistance |
|
| 51:02 | non uniform memory access. So the particularly on the servers you typically and |
|
| 51:15 | bridges to step up to the two service as we call it. So |
|
| 51:21 | memory in the service is associated a subject. So it doesn't mean that |
|
| 51:29 | course and one socket cannot access the on the other socket. Yeah, |
|
| 51:37 | it's further away so the time to the an item defense among other |
|
| 51:49 | the relative rate or replacement of the and that I died whether it's in |
|
| 51:54 | memory on the same stock belonging to same socket or the other side. |
|
| 52:00 | so systems nowadays they have also features they always try to help you to |
|
| 52:08 | placements and excess pattern. So predominantly on at least the frequent access is |
|
| 52:14 | local stuff that's supposed to what's on other side. So that's what the |
|
| 52:19 | features is that there is ways troubling ends up in popular system. So |
|
| 52:27 | again, if you failed to exploit performance was down and let's see what |
|
| 52:34 | things. Yes. So uh, you have stuff like things, things |
|
| 52:44 | to consider so if you don't have one access but some other for arbitrary |
|
| 52:50 | parents remember violence is considerably okay. huh. Then the animals to be |
|
| 52:59 | . That screen isn't feeling the performance is used as the member of |
|
| 53:06 | But the big performance is really just at the Corporates of the memory models |
|
| 53:14 | a number of its second read or and the single cycles and that is |
|
| 53:18 | pete memory that the state what's actually of the street that is used |
|
| 53:26 | So this is kind of just seeing aspects oh, the call almost can |
|
| 53:34 | you a more realistic do once achieved then use cancer rate your home full |
|
| 53:44 | against. They sort of scaled down depending on the properties of the uh |
|
| 53:55 | . Just a sign. I already it on the other side, but |
|
| 53:58 | just going after the specifications of memory , CPU's and multiply numbers are Oh |
|
| 54:10 | and so there's things taking watch. why don't I think just to them |
|
| 54:19 | then they haven't been successful, but to get to that or not. |
|
| 54:26 | problem is useful. All right. easy if I share my strip. |
|
| 54:44 | . Okay. Yeah. Uh it's using to shut out footsteps. |
|
| 54:51 | yeah. So any questions on those and what it's very frequently used to |
|
| 54:58 | it person. It's a simple modern , it helps you sort of |
|
| 55:05 | first order of business trying to figure that this member about replication. I |
|
| 55:13 | to focus on those students. okay. Oh, I'm prepared |
|
| 55:23 | you know, putting it over there they can see it as well. |
|
| 55:29 | , right. So again, we the yeah, the timers in the |
|
| 55:41 | , but just quickly. How, you can use them in your in |
|
| 55:45 | goals. So the first one the time of day 1? Uh |
|
| 55:55 | Okay. Okay. Yeah. That Yeah, yeah. For uh, |
|
| 56:09 | at this time of day one again has a microsecond level uh resolution here |
|
| 56:17 | when you can use and then forget . Okay it's it reports the world |
|
| 56:23 | time. Uh so you need to careful uh in case you have an |
|
| 56:29 | that will be using uh we have idle media started waiting for I. |
|
| 56:34 | . To be finished. So and is the third time of they may |
|
| 56:37 | be may not be the best one use. There is a successor to |
|
| 56:43 | time off day which is called the . Get time. And that allows |
|
| 56:47 | to use a different type of cloth I'm just showing your two of |
|
| 56:51 | But if you go to the documentation I'll let you she was at least |
|
| 56:55 | or six different class that that you use. Uh When is this called |
|
| 57:00 | ? Manana ethnic? Uh It reports the wall at that time. But |
|
| 57:05 | good thing about it is that it implementing every time. So it doesn't |
|
| 57:10 | on if the man has changed any already, huh? Has messed up |
|
| 57:15 | the with the system. So it's that two successive all the time. |
|
| 57:21 | please be larger than the first And with uh if you use the |
|
| 57:30 | process, if you time I. flog this one here uh that will |
|
| 57:35 | will give you the city your time your processes. So if you're if |
|
| 57:40 | call a function lets you sleep it's gonna be sleeping for uh for |
|
| 57:44 | one second periods, it's not doing anything. So when you use |
|
| 57:50 | c p time I got here, was it will report a little bit |
|
| 57:54 | time because there's a little overhead and the sleep function. But if you |
|
| 57:59 | a large function that uh maybe maybe waiting in car, I ought to |
|
| 58:05 | finished and you're you're only interested in amount of time that it was actually |
|
| 58:11 | something. And then uh you might uh will be the one to go |
|
| 58:18 | , right? The 3rd 1 that saw is the law under 40. |
|
| 58:24 | this one actually, again, it you the CPU time again, but |
|
| 58:30 | was a little bit, little bit . Yeah. And it actually doesn't |
|
| 58:37 | at the same frequency as the CPU . So there's uh it uses this |
|
| 58:43 | for second. Valuable. That's implementation on every system. And the way |
|
| 58:49 | works, is it the collateral second usually uh on the on the assembly |
|
| 58:56 | is one microsecond. So everyone Microsoft the uh increases. And that's why |
|
| 59:02 | you compute that, compute the difference blue balls for the clock function, |
|
| 59:08 | it's not going to represent the actual frequency clock cycles, it's gonna be |
|
| 59:12 | gonna be a multiple of the philosopher second. And just quickly show how |
|
| 59:17 | Alberta was for them so far as uh don't look at that time. |
|
| 59:22 | here I only said sleep function for seconds. So when I use the |
|
| 59:27 | clock it gives you uh these many seconds which is over five seconds |
|
| 59:34 | And as you can see the city basically did nothing. But it was |
|
| 59:38 | but in a lot of the clock computer the whole five seconds. But |
|
| 59:43 | you use process CPU line club, a slide overhead involving the sleep function |
|
| 59:48 | then coming back from the sleep So it only uh computer that much |
|
| 59:53 | of time when the function was actually and not a leap period there. |
|
| 59:58 | that if you use another dog with at that time, that's the real |
|
| 60:03 | time on a trip to the whole second period for which the process was |
|
| 60:09 | . Yeah. Uh And then the uh again these texts, they are |
|
| 60:16 | representative of the C. D. . Or pregnancy. Uh the multiples |
|
| 60:22 | uh this 102nd resolution here. And when when you try to computer, |
|
| 60:28 | using the start and end with floppy divided with uh proper second um |
|
| 60:36 | And that gives you the structural alignment . Just stimulating the difference between these |
|
| 60:41 | is not representative of the people of here. Um The other one other |
|
| 60:50 | was so again it represents the walk time. And this is just basically |
|
| 60:56 | Um 1 1 million something operations are . That was in the code and |
|
| 61:02 | , so that represents the world of . So again, if you have |
|
| 61:05 | line in your uh uh don't try use that any questions about the alzheimer's |
|
| 61:16 | a very kind and wonderful. Mhm my next one is uh that guarantees |
|
| 61:23 | it's gonna be consistently increasing. So so many times that uh huh or |
|
| 61:34 | admin uh the administrator of the clark the time. So they uh consecutive |
|
| 61:43 | and then it's likely that the uh 2nd maybe maybe less than less than |
|
| 61:49 | question one is something that the crop without monotonic it doesn't it doesn't get |
|
| 61:54 | by such factors. So it's going be always increasing. Oh goodness. |
|
| 62:04 | it doesn't really depend on actually you know that you are on the |
|
| 62:16 | program and so as I don't know . So at the same time |
|
| 62:24 | for for two to diversity is very like uh cause for both trends may |
|
| 62:31 | different values because the the execution beginning each track may be different. So |
|
| 62:37 | starting point and and union point will different for each the uh I do |
|
| 62:43 | believe they share the share that if remember correctly. So you can see |
|
| 62:51 | of the for time from the impacts the for this process of your time |
|
| 62:58 | . I think uh that's what I all the shots and in the process |
|
| 63:04 | another club called thread CPU time that is safe and if run if used |
|
| 63:10 | a single thread it will uh return reliable time for each particular threat that |
|
| 63:17 | that function. All right. Uh time. Yeah. I'm not very |
|
| 63:33 | right. I'm working. No, goes on in the system in this |
|
| 63:39 | . Fine. Are they all the ? It's always wonderful to five |
|
| 63:49 | Mhm. Usually are very well connected something. And the other naked find |
|
| 64:02 | . That's the early calls. There one I would say it's all about |
|
| 64:11 | experience and I think there are more committed on the basis for it's supposed |
|
| 64:18 | be an or because one of them a democracy helps unless just always hold |
|
| 64:27 | off. Oh there you are property markets are very often that's oh get |
|
| 64:44 | here. Yeah. So you can the parliament by using the article but |
|
| 64:54 | enough. I'm also a particular Yeah. Usually yeah. Yeah. |
|
| 65:04 | huh respect. Uh huh. Oh yeah. When they're doing that |
|
| 65:13 | truly have exclusive. All right. . Yeah. Following off. But |
|
| 65:26 | don't I don't believe there uh either the sites are giving exclusive access is |
|
| 65:31 | days I don't think either of the they are giving exclusive access is I |
|
| 65:35 | to I tried to go through their license from confederation all Yeah, They |
|
| 65:47 | to go on Bridges one. But the registry. Yeah. Yeah on |
|
| 65:55 | bridges is used to work but now bridges to and stamp it to |
|
| 65:58 | None of them. Mm Okay. let me go through the first |
|
| 66:13 | So yeah so you can hear the of scream. I'm sure uh any |
|
| 66:20 | you may have already uh done it few times just to give an idea |
|
| 66:26 | . Um when combined for using one elements for for each of the each |
|
| 66:33 | the areas of stream, it declares areas in the world. And so |
|
| 66:39 | total memory required here as you can is uh close to 23 MB which |
|
| 66:44 | still less than the year. Uh recently took a bite on this uh |
|
| 66:52 | these notes. So as you can here the data is that you get |
|
| 66:56 | all the four functions Uh close to device for a second to 22 device |
|
| 67:03 | a second. And this is why you're in the big uh regions. |
|
| 67:09 | let's say when you go 200 million and that's obviously quite large than the |
|
| 67:18 | the LP gas. And it's very that most of your accesses will come |
|
| 67:22 | leader from the main memory. And you can see uh a huge drop |
|
| 67:27 | the uh in the in the band here. So for try and it |
|
| 67:33 | from 22 MB/s to Down to 13 . You can see the difference you're |
|
| 67:42 | to access most of the real compassion the main memory. Mhm. And |
|
| 68:01 | that's the that's the largest one I at least. But yes the first |
|
| 68:06 | is actually busy so I already learned schools may not get access to the |
|
| 68:10 | quickly dot com. Mhm. Um then just quickly with God's here, |
|
| 68:18 | drops Uh here takes three Parameters. first uh So the first one is |
|
| 68:26 | size of the table. So if live 30 that's to cover 30 side |
|
| 68:30 | . Uh uh Look at nuclear elements the in the table and the |
|
| 68:35 | Uh I'm interested to find how many these are performed on was uh enabled |
|
| 68:40 | each process. Yeah a second and you need a banquet per second. |
|
| 68:52 | with the two billion elements which are likely that will be accessed mainly from |
|
| 68:57 | main memory. You have to give the per second rate uh those 2.06 |
|
| 69:03 | . But if you come down to just uh one megabyte elements between our |
|
| 69:10 | here that right Uh increases the updates 2nd A 2.32 again. So uh |
|
| 69:20 | accessing data from past versus the time again if you don't bear stream end |
|
| 69:25 | stream accesses data manner again doesn't have access to china stride one manner. |
|
| 69:34 | that's okay. Okay these I I remember but I think these are 64 |
|
| 69:48 | numbers. And so they sent, out there is a reason, right |
|
| 69:55 | apologised. I heard this name also . The 1%. Uh huh. |
|
| 70:22 | . And so Dije here uh little and I'm going to talk about is |
|
| 70:29 | we thought uh it stresses the computer ability of uh of any system |
|
| 70:36 | So in that case what you get the flow rate and that represents the |
|
| 70:42 | the president of the components of the course here. So for this year |
|
| 70:47 | I used uh anti vitamin K elements are again double precision. And so |
|
| 70:53 | you uh compute the total number of that do and you and divided by |
|
| 70:59 | amount of time taken for the whole of the islam. Uh I'm close |
|
| 71:05 | 95 years or less per second. again if you go with smaller smaller |
|
| 71:10 | are slightly that they will fit in past and then you see the performance |
|
| 71:15 | increase uh at least a few hundreds times. So again there's a difference |
|
| 71:21 | uh these 33 benchmarks screaming upstairs just memory systems and uh this is the |
|
| 71:29 | with the ability of the of the there. And again when you when |
|
| 71:35 | profoundly experiments the different sizes of committees you see the difference in the |
|
| 71:44 | So he won't talk about you looking uh you want just uh oh |
|
| 71:58 | sorry about it. Right, Yeah so I guess Want Children to |
|
| 72:17 | top 3000? Yeah, convincing. that way yes. Mhm. So |
|
| 72:32 | a cover on uh on this, went on the left here. So |
|
| 72:38 | anybody see any problems with this uh court before I get into the |
|
| 72:53 | Mhm. How many? Uh What okay if anyone doesn't see any |
|
| 73:17 | ? Let's see if when we come , what happens? So they refuse |
|
| 73:23 | gcc come by right here. So the school does it it uh allocate |
|
| 73:29 | memory for this offer and then stores string that you enter from the console |
|
| 73:35 | the city and into the end of buffer. And finally Princess, so |
|
| 73:42 | see what happens when we run So they said if I just entered |
|
| 73:45 | string, just Uh huh It actually given recommendation for and now just by |
|
| 73:53 | looking at the score, I don't anyone can uh say that. What |
|
| 73:58 | might be the problem with it? the way to deliver is uh in |
|
| 74:03 | on violet with a flag gcc. and then uh she and then again |
|
| 74:11 | simply where you would uh compile Yeah. And what having the minor |
|
| 74:19 | that does it it actually preserves the tables that when the compiler compiles the |
|
| 74:25 | so that the people can use it uh to see through what might be |
|
| 74:29 | on in the cold once your combined the minus the flag, what you |
|
| 74:35 | do is use the command GDP. uh just run your um executed with |
|
| 74:41 | GDP to cover and that happened that a GDP council here. So is |
|
| 74:49 | The last one is Mhm. Natural look. Yeah. Okay so |
|
| 75:09 | you once you're in the GDP console you can use the command run uh |
|
| 75:16 | on the run the program with the there, that's okay, three sides |
|
| 75:22 | it over more. I was just . Okay. Yeah. Okay so |
|
| 75:36 | let's hear let's hear around the program in the report earlier again when we |
|
| 75:42 | this uh it tells you that it examination here, how everyone coming |
|
| 75:50 | Oh that's why one. Mhm. there we go. Okay so in |
|
| 76:15 | this debug responsible here. So we the program, you know we can |
|
| 76:20 | that you were also turns with this whatever now inside the console. Before |
|
| 76:25 | get into the debugging uh part of part of the demo. Yeah you |
|
| 76:30 | use that on on health and you and you can use all sorts of |
|
| 76:34 | command categories that are available for you use insight as you begin to |
|
| 76:39 | So it allows a break mine uh a program checking the back another program |
|
| 76:45 | other uh different features are there that might use when you're doing much more |
|
| 76:51 | because look in here. Well it's simple even in here, the first |
|
| 76:55 | that you can use is that's called batteries once you once your program is |
|
| 77:02 | and what the mattress programme does, shows you the stack trace of your |
|
| 77:07 | . So it basically tells with their functions or call there will be uh |
|
| 77:13 | . Now these uh whatever this in and I guess these are in the |
|
| 77:19 | for uh percy libraries. The function many got a call from our source |
|
| 77:25 | was the main function. Now these here on the left, they represented |
|
| 77:33 | frame as a reliever because it and we use frame command with the number |
|
| 77:41 | with your with the function that you're in art is the main function. |
|
| 77:46 | tells you that we're exactly in the the cold the record. now, |
|
| 77:52 | tells you that it happened online. here, which is this one here |
|
| 77:57 | the afghans function. Now. Generally going to learn that the uh see |
|
| 78:05 | bed functions like f gas or any uh in the way it was like |
|
| 78:10 | syria and they're uh they should not any problems. So the likely reason |
|
| 78:17 | is one of the variables that media offered and what buffer should um should |
|
| 78:24 | had but many years maddog with It should have had uh memory |
|
| 78:31 | So how do we check what's inside her here? So for that we |
|
| 78:35 | use the command here and get the of the variable that we want to |
|
| 78:41 | the contents of. And when we that We can see the buffer actually |
|
| 78:48 | a 0X. You know, has small value which is just basically another |
|
| 78:51 | here. So what happened wrong Ramada should have given it a memory |
|
| 78:58 | right from the and in England. . So let's see. So first |
|
| 79:06 | union capital program here and what we now do is make sure that my |
|
| 79:13 | is working correctly or not and to that. You can use a great |
|
| 79:18 | . Yeah. Uh simply great here provide the source code Violin as well |
|
| 79:28 | the line number on which you wanted set a break fine. And what |
|
| 79:34 | going to do is it's going to a breakpoint on that line. And |
|
| 79:38 | next time you run the program in quality control it's going to cause the |
|
| 79:43 | of that that line in the cold now uh what's what's the above? |
|
| 79:52 | able is getting inside from the mouth function and you can check here |
|
| 79:57 | you can see it as um as value which is an old value. |
|
| 80:01 | there's something wrong going on going on the with the malfunction. Now, |
|
| 80:06 | you can go injecting malfunctions documentation, know that when malfunction cannot, I |
|
| 80:13 | get the space that you requested it generally returns a null value. So |
|
| 80:18 | one last year of 31 a period for too big a commitment from the |
|
| 80:27 | , which is very large and it's that the system and I gave |
|
| 80:32 | So let's see what we can Let's see if we try to reduce |
|
| 80:36 | amount of money. So let us it magnifies are open to advice. |
|
| 80:42 | save it and I'm out of the uncle here using the command q use |
|
| 80:50 | could uh in the uh GDP Both uh my baby. If you |
|
| 80:57 | buy the goal here again Now without uh 65 because we don't want to |
|
| 81:02 | it right now innocent engaged here that . And then and it works just |
|
| 81:12 | . So this is just a simple flow workflow, pinpoint the problem |
|
| 81:18 | your goals. And so and one the assignments will be providing couple of |
|
| 81:22 | where you need to find where the is. It's not going to be |
|
| 81:25 | complex. But this is just one you can try to pinpoint where the |
|
| 81:29 | is in your good and rather than to print us every place, see |
|
| 81:34 | in the innovative alarm on and that's the people are maybe very useful for |
|
| 81:40 | in that. So this is not my debate to do it. Just |
|
| 81:46 | of the way to do it. , sir. Yeah. Unfolding |
|
| 82:05 | I don't think this was it. . Questions. Scientists are our |
|
| 82:27 | But that's that's good. Soften Very point okay. Yeah. If |
|
| 82:43 | have a 2009 code you don't want be doing print us and that we |
|
| 82:49 | be really helpful. Yeah. Uh Right. That's like the compliment. |
|
| 83:03 | sit down. Yeah. Yeah. . Oh, any other questions otherwise |
|
| 83:21 | finished controversy. Very good. Oh . I want to ask you one |
|
| 83:27 | the best part of code for hard you may already have multiple expectations so |
|
| 83:34 | have to run again. How awful ? Yeah. Oh, terrible. |
|
| 83:46 | . So I don't, uh, don't remember what it looks like. |
|
| 83:51 | has already looked at your parameters. number of times that runs away after |
|
| 83:55 | looked around. No. So at if you want to try different sizes |
|
| 84:01 | the race. Yeah. Then then need to read them violet every |
|
| 84:06 | It doesn't have a loop around. . Right. Our system. |
|
| 84:12 | but and you need to pass the every time for the data sizes. |
|
| 84:19 | , Unions are running three times for same parameters. I just have to |
|
| 84:25 | friends. Yes. Also before his question. So I don't think this |
|
| 84:47 | going to really quick. This wouldn't for most part it's just a few |
|
| 84:53 | of execution for the society correct? might say something encouraging. Islamic mystery |
|
| 85:04 | , preferably five times as long as it's just a second. You |
|
| 85:13 | the drums are more or less this tests. You know, hopefully |
|
| 85:26 | you will kind of be lucky in sense that it reported times and |
|
| 85:33 | Very little variation but it could also outside and then I guess it's |
|
| 85:42 | you know, that can happen in principle that should be much wanted. |
|
| 86:05 | there's not anything once is and then risk in the transit. No, |
|
| 86:13 | really does some times. Again, else happens in the system in particular |
|
| 86:21 | to show at the moment, it be awfully significant. Long term. |
|
| 86:27 | totally are. Yes. So another all got accounts have tried use the |
|
| 86:55 | and life. Oh, up to , I've noticed that not from |
|
| 87:06 | but over the last, so this two systems, the sun is relatively |
|
| 87:13 | to the half years old. Unfortunately messages that it's down, everything didn't |
|
| 87:24 | trouble with it. So on the hand, stand please in something very |
|
| 87:29 | . It's an older system and that's . Good to start this campaign. |
|
| 87:37 | things have stabilised roses too. They're a pretty good and bring it back |
|
| 87:43 | again. Quite clear by this guy something more. That means an office |
|
| 87:51 | and a few times usually answer one back is not good. I'm sorry |
|
| 88:00 | that good system. It's not stable stampede even though it's stable but the |
|
| 88:09 | times have long. Yes. Again popular. Yeah. Coming back to |
|
| 88:16 | question for trying to get multiple It's a good, good thing if |
|
| 88:21 | get practice to writing scripts, although of that was that ultra wide. |
|
| 88:25 | already have scripts but try to write that automate your jobs. Jobs and |
|
| 88:31 | and getting all the outfits. Because you don't want to be sitting |
|
| 88:36 | , we didn't get a chance to . But as you eat these simple |
|
| 88:41 | are identified. They don't have to complicated but just enough to get your |
|
| 88:53 | . Yeah. And also I think this for this first exercise but trying |
|
| 89:02 | maturity generate outputs in the format that can also processed by machine and all |
|
| 89:07 | of everything. Yeah, there's some . The tables, tables trace some |
|
| 89:21 | unto themselves. It is for me graphing, make things stand up more |
|
| 89:30 | you want to take it off. want. The difference is probably numbers |
|
| 89:35 | was asking shows it's very clear what something is anomalous and mm mm we're |
|
| 89:44 | transit checked instance. Someone thinks it's good to try to also things about |
|
| 89:52 | and on process for trying to use help extract kind of real information without |
|
| 90:08 | , okay. Yeah that is And this time phenomenal tools you want |
|
| 90:20 | use analyzing what things that actually Oh, that's what's doctor talking? |
|
| 90:46 | mm Okay. Yeah. Uh Yeah. Well, I think shoes |
|
| 91:12 | . Oh that's uh no. |
|
