3: Sysstat Demo

This transcript is also available as a PDF download within the files tab.

Sysstat Demo

Now this is what I’ll do in this demo. First I have to open Linux terminals from which I can work. Then I have one cluster chilled terminal, and then on Linux I will generate io to an NFS mounted volume in the costume. Then from the Nuccio in old tab, I will monitor the consistency point types and the command that I will use is Sysstat, dash you and the interval is two seconds. Now let’s do it. So I’ve got my two Linux terminals at the top and my clusters here at the bottom I’m going to be using storage machine SVM blue, and there’s a volume which is called blue and Granicus. It is in node one egg one it’s two gigabytes in size and I have some space left.

Yeah, so at first have a look at the contents of my volume. I’ve got 367 megabytes of data in there. Now I go to the note show of node one because that’s where the aggregate is and the volume and I run the [inaudible] command [inaudible] there’s you and two seconds we’ll be the interval. So I look at the consistency point types and uh, so I’m going to follow all of these columns and see what happens. Uh, forget about this time consistency point. That’s because of the DEU command. Uh, reading files also means to writing access times now, but for the rest of this cluster is very quiet. So it’s only zeros, there’s no activity going on. And let’s create some, uh, I uh, uh, I’m going to run a small loop in which I create a tough aisle with the name of my Linux machine. So the tar files going to be called Linux one, and the time command will register how long the command will take.

After that I removed the tar file and the whole procedure will repeat. So, um, basically I’m just generating io and what you instantaneously see is that I have a full consistency point, a reason the type is full, meaning my envy ram is half full and also hash. Mark tells me that the next consistency point will be a back to back consistency point. Also, you see that the consistency point takes six or eight seconds to complete. So this system is overloaded. Obviously it’s only creating consistency point. Let’s add another loop from the second lytics machine and that will really bring down the system. So it’s really suffering.

It’s only consistency points. So we have a problem performance wise, you see that the time it takes to create the file will increase as well from 24 to 28 and that will only get worse. So it’s 41 seconds now and my NFS client is even getting a steal NFS handle because of lack of response from the, so let’s stop the suffering and finish the loops. And after we’ve done that, uh, the system returns to be in quiet again and we’re done. So what we’ve seen is that what four is more than just the file system. That data has to pass three software modules before it can be written to the volumes and the support modules are networking scuzzy plus this issue manager and the data module. Then data will remain in envy. Rem until a consistency point is complete. There are different types of CPS that we had to look at and back to bed. Consistency points indicate performance problems. If you have any questions, please let me know.