A free video tutorial from Jigar Vora
Planning Manager at Panasonic
3.9 instructor rating • 8 courses • 30,398 students
Learn more from the full courseTERADATA: Data Warehousing technology
Teradata tutorial with complete real life examples for people who would like to get certification as DBA.
06:07:02 of on-demand video • Updated June 2015
- Understanding Datawarehousing concepts
- Get Certified as : Teradata Certified Technical Specialist
- Get Certified as : Teradata Database Administrator
English [Auto] Hello folks let's take our discussion forward towards started it up architecture. So in the last screencast I talked about like a massively parallel processing system and S&P symmetric multiprocessing system great. Now let's talk about the for the architectural blocks of it. So if you can see like this is a machine now and when we find a way the it could go to these chandlery drivers get it I get it so I'll explain that with an example. But before that just let's get acquainted with these terms and then spin that one example. OK. So as an inventor we had a query. So it passes through a certain target we channeled travelers and then we have something called a B which is what you call it as a passing engine. OK. And then there's something called a by net Traver. After that we have him be accessed Mordieu processor and then there is a virtual disk. OK. So the query that you submit passes on to the teller data gateway it goes through the parsing engine then by Nate and braver then MP which is a virtual processor and then into the virtual desk. OK. These are all things the passing engine the binary driver and then the Impey is contained in something called Zappy. OK. So tell us where the land gate gateway and the channel traverse are further and as processes so these things will run as processes the channel over the gate when the oil leak does get with them as processors and then Impey and passing engines other virtual processors the processor which will run under parallel database extensions so Pudi is something like an operating system so you'll have a Windows operating system on top of which you can install multiple office right. Similarly you have a Pudi which is like a software on top of it which we have this parsing engine. This impedes this blindness. OK. And as Apple in the last session that a single or which just contains 1 Nord will be called as a symmetric multiprocessing system or if you have multiple nodes we call it as a massively parallel processing system. OK. So taking our discussion forward in terms of what is a better way to get to the channel driver what is a passing engine. What is a binary driver. What is an MP. What does a virtual disk. What we mean by a PDU let's understand these example. So let me get back to max of shit OK. So I was talking about this this stuff that we later Square-D we execute liquidy and we get to set up. So let's take this one as a starting point and discuss it further. So as soon as data ready it passes on to these deaded does get we look at as soon as we write a query that data it passes to the elevator get. Query to pass through this gateway which is just an entry point of score who that and get it to pass what did it take to we then do have something called parsing engine passing is something which it is possible to do my part that's firstly as soon as you wrote the as soon as you'd order the what happened happen next. You need to log what it means to them. Right. When they say you need to logon to a database to do an acquittal it means that you need to be authenticated by some system. Suppose you are trying to access a database. That database will track what establish your credentials straight whether you were like a legitimate user or not. So the first task was the parsing engine performs is authentication it to see whether the guy who is submitting the query is the unknown person are not allowed to do that or not. So once that connection is stablished ones that like authentication is done the next step is to see that as well query with shoot or where you might have some Eddas state your query you might have made some syntactical errors or some logical error. There can be a number of errors in it. So the second thing with the parsing engine does is to parse or is it when I say parse it is Cuil it means that it will check your school ready for syntactical issues. If that is had technical issues on it or if there are any errors the parsing engine will drop a message back to the delegate I get it and back to you that you have written erroneous. So you need to get this Quillen submitted back. So this is the second step or second thing with the parsing engine does no indication it establish the passing engine has seen that what is that. Yeah it is working fine. It doesn't have a syntactical issue. Then the third thing what is done this optimize that optimized liquidy not assume that the US had like an MPP in RAM and massively parallel processing environment. You have multiple processors available but you have a proc one proc to proxy these three prox that are available. So the optimizer in passing in general decide that this particular query should be given to which all processes are set. OK so what is the most optimized way of fighting the desert out of this square. Should I just give it to one or. Or should I give it to naught. Or should we give it to all the north so that the query it will becomes faster. So this decision on how to optimize your query how to get that result in a very short interval of time is something which is decided by the parsing engine. All right. No one said that it is done syntactical checks for this is done. It's optimized. A plan has been set up there. OK let's follow this plan. We'll just allow that to be passed to these two processes because that is the most efficient the most optimized way. What is the optimal optimize optimize plan is prepared the next step is to boss on that plan. The next layer you know the next list is before that to pass on the plan to the next so we call it to the dispatcher. So there are four things which are done by a passing engine. First it authentication. Second is passing year as well. Third one is optimizing the query so that it will give you an optimized output. And the fourth part is that despite that plan other optimized Banbridge you have generated here to the following layers. All right so now what is the next layer. So what's your goal. Where are you past. You have gone through the data gateway then to a passing engine. The next step is called Azer bidon it. Look at Beinecke is something which is responsible for handling the traffic in your machine. One minute. So you have a buy in and it's you to buy that one. So by now it is something which is just acting as your traffic control tool. So what generally happens in a traffic control tool like once you're when it is submitted to what I get and then the parsing engine and that particular square is being executed and support instantly a second developer he's brought in one more quickly so that particular second query really be on hold until the first query is completed. First Square-D has been executed successfully. So this decision as to whom Bouvard or which Square-D to execute or how much time to read all these balancing act so that your services which are below doesn't get affected all the Lord balancing act or traffic control act is taken care by by next know by next act when number because traffic control or balancing something which is of utmost importance. So the architecture has to buy Neff's just for the simple reason that if one buy and it goes down it becomes out of order then you can have the second by next which can take a few steps. OK so just forward balancing balancing means or Lelo those which can be processed if the other is can we just keep them on. Once we process one set of what is then we'll good with the next set. So that is taken care of by the traffic control pool which is by net zero and by net 1. So we can call this a balancing act or traffic control tool. OK. And then why we have to buy an independence you don't buy it one way because if one bilinguals don't we can have the other by and by not to rely upon. So these two by nets provide fault tolerance if one was at fault. The other one can take care of the stuff so we call it a forward tolerance mechanism. OK. Now we have processus work we call it as in these Okay. Suppose I just have three or less employees. These are virtual processors. The prox these are not physical entities. These are logical entities. So we have broad also virtual processor what we call it Access Module processing and we call it DOES IT imbue one in three to this one amp three. So you have three amps which like the plan which was generated by the dispatcher here. It provides the plan which is generated by a passing engine to a player. And then this rock so we know right if you have a process you need to have a hard disk also in our computers. So similarly if you have a rock you need to be this. So will have these VDX So this process will have something called the disc where you could actually do well at the site which are disk is the place where you are and you will decide who will be managing that data who will be helping us to face that data. It is the employees which we've to the virtual processors to Abai net which acts as a balancing layer from a passing engine which a lot indication passing optimizing and dispatching and then it negatively. So once you write a query these are the things which happens when we see the equities getting executed. It's not as simple as just pressing a five candidate and getting out. But these other things to which your entire Kuwaiti will pass through. And this is what cost is your execution. OK. So the lady it passed on Bill that added it to me through Opah singing which did all these for us. Then you pass on to well by Nick what you'll do a lot. Balance the activity and you have by nurse to provide for her tolerance from. Plan which was generated here. OK optimized plan which originated here will pass on to your employees. Employees will exit Judo's plan business plan it is hard to store the data. Execute those plans and restore the data on this much does. So this is what we mean when we say like we're executing a query in the data and this is the architecture of a tentative date system. OK. And of to be executed once we execute the plan once we get the data which would be submitted that itself back to the output layback. They use it to work at it again. So again once you get that suppose the requirement was to get the maximum salary to go to all these years once it gets the maximum salary the maximum salary output will go to this MP then to buy it and then we're passing and then and finally to the terror target. So while the data is getting stored it followed the same levels and while the data is coming back to us form of onset it could follow the same flow from the view this to your imbues from the impis to the By next then to a passing engine that ordinarily Pegatron. So but it also can be closer to the transport protocol. It can take the data from parsing engine and dump the prop. It can take the output from the view broken down into the passing engine and I get it. So this is what is the architecture of better data. OK. No I just have to stay where we are using a parsing engine. You have a buy in IT employees and the word sordidness. OK. So the session can do so late. It checks for all tradition and all will be taking care about the parsing engine which pulse your data optimize it and then dispatch it to the corresponding virtual disk or the user. OK. And one more important thing is that at present parsing engine can support up to 120 sessions which means that if you have a standard the architecture then I don't own one. People can submit the Kuwaiti at one go. And the buzzing engine is that much capable that it can run all these 120 sessions palely through this Biden by now. So by now it can act as a trance like Lord balancing. It can take one going two sessions at one go and can distribute among different IYAM and the virtual desk to work in parallel and get to the hour. And this is what is something which differentiates stared at it from the rest of the doods because it can do us in a massively parallel processing way to these basic building blocks of the architecture. So this parsing engine can take up one sessions at one go which is huge. So one when guys can find Eskil is at the same time and this by and it is also that much powerful that it can balance the load among these imbues in a very soft money doesn't require any other lake external factors. So by net is self-sufficient will take care of one Quinby sessions which can be supported by parsing engine and then for the impis can work in tandem and can get it done. So in the MPP environment parsing engine bind their consolidated with an MP and a virtual this mix up then Benedetta architecture. OK. And the data be so profitable to passing and then by in the process or not. So these are all virtual processes. So to run a virtual process or to run a software what we need we need to have an operating system. So these things are impedes our parsing engine by net. All these had an owner software what we call it yes you'd be good. So if you could see here the parallel database extensions is something which would take care of your parsing engine in piece and the bind drivers. So all these since we're all entities these are logical concepts so that they need to be on top of the operating system so you can imagine that a little database extension PDA as an operating system which you'll haul all your virtual processors and it's OK. So this is what we call it as the architecture of data to the important points that your passing engine or the task at perform it performs that authorization session control parsing optimizing dispatching then you'll have a net driver which will take care of a lot balancing activities. Then the plan which was prepared by the passing end will be passed on to the imbues which will store data on the different virtual this aspart the plan generated by the parsing engine and all these are put on top of a pile of little database extensions and we create our systems to target it which can be an Escuela system or anything. And this is how you get into liquidators. OK. So this was all about an architecture of. To take our discussion forward on different the technicalities associated with that. In the upcoming scheme costs. All right. Thank you guys.