Recently, a database manufacturer said to a customer that the performance data they got from a customer was much better than that of an external storage LUN. The customer said that since the understanding was inconsistent, let's go to the customer side to make a clear analysis and draw experts from the database manufacturer, so I also made a special trip for this. At the meeting site, the customer said to cast out the performance data of the test. The database eldest brother finally got the data from the brothers who changed hands many times and cast it out for the customer to see, it is exactly the same as the data in my hand. See the table below.

Then, the customer sees the two tables, one is multi-concurrency scenario and the other is single-concurrency scenario. Obviously, there are many concurrent scenarios, regardless of latency, performance data, the performance of external storage is much higher than that of local disks, and the latency of random hybrid reading and writing of local disks is as high as 6ms. When the customer sees it, he says, XX, what you said is wrong, it is obvious that the external storage LUN has better performance and latency, which makes biangbiang face-to-face. However, the database brother keeps saying that in this single concurrency scenario, the local disk is better than the external storage LUN in sequential reading and sequential writing models. The customer directly interrupted that the transaction database is mixed read/write, where there is full-order read sequential write, I am sure to see the performance data of multi-concurrent mixed random read/write scenarios, moreover, a single concurrent hybrid random read/write is not as good as an external disk. The database elder brother didn't know how red his face was. He kept talking about the good performance of reading and writing local SSDs sequentially. I really didn't know where he got the courage to open his eyes and talk nonsense.

Another database expert found that it was not coming to the station, so he interrupted quickly. Let alone, we are not familiar with this test data. In fact, our database is separated by a general-purpose OS, and we don't care whether it is an external disk or a local disk, only know the OS block device drive letter, regardless of the disk, do not know whether it is a local disk or an external storage LUN, of course, use the storage LUN, multiple disks for RAID and slicing pool wear balance, the performance is definitely higher. The FC high-speed link has low latency, and the most important thing is that the storage reliability is higher. If the local disk is not RAID (hard RAID card is also a single point, which increases the latency loss), if the disk is broken, the node will be switched. If you use an external storage LUN, reliable RAID guarantees that the node will not be switched. We definitely support external storage.

Later, the database expert spoke human words and praised them. I also expressed my gratitude. In the future, I will strengthen cooperation to save and supplement the database shortcomings with high reliability, because it is basically impossible for a database to be highly reliable, tens of millions of lines of code are extremely complex, and coupled together, it is even worse to do a database. However, the bricklayer in front of us really didn't know where the courage and courage came from to talk nonsense in front of the data that the customers looked at. What was the professionalism of such a person?! A confrontation ended happily. I thought that brother was so embarrassed that he didn't know if he wanted to find a hole to get in? Return to the essence of business, seek truth from facts, professional people do professional things.

Source: Bi xunshuo