The reason that the server went down on the first day of Amazon Prime Day is 'the reason why it was changed from DB to Oracle from Oracle'

By Prohibited Network

At Amazon Prime Day 2018 which started at noon on July 16, 2018 in Japan time, the server fell down immediately after the sale started and the access fault occurred for about 1 hour. According to a report created by Amazon's internal survey, this problem is increasingly likely to be the biggest reason to change trading database (DB) from traditional Oracle made in-house.

Amazon move off Oracle caused Prime Day outage in warehouse
https://www.cnbc.com/2018/10/23/amazon-move-off-oracle-caused-prime-day-outage-in-warehouse.html

Amazon Prime Day 2018 established a sales record that is "the highest ever" for the company. Amazon is known not to announce sales, but in 2018 it is said that more than 100 million items were sold during the sale period, and it seems that there was sales of 200 billion yen in Japanese yen It is.

Amazon, sells more than 100 million items with record high sales on prime day - iPhone Mania
https://iphone-mania.jp/news-219376/

It was a record that made us realize the worldwide "net mail order top form" again, but in reality it is clear that the server was down on the first day of the sale and opportunity losses of 10 billion yen or more occurred in the Japanese yen It is getting.

Even Amazon Prime Day recorded the highest sales ever Opportunities loss of over 10 billion yen in server down - GIGAZINE

According to the CNBC report that Amazon's research report was obtained, the obstacle that occurred in Ohio's largest Amazon warehouse (fulfillment center) in Ohio occurred as a bottleneck in DB processing capacity It is said that it is being analyzed. Amazon has set a policy to replace the product management DB with Oracle from 2020, and many factories already introduce Amazon Aurora PostgreSQL (Aurora) of Amazon Web Service (AWS). However, it seems that the whole system went down as the new DB ceased to correspond to the explosively increased order quantity.

One of the factors that caused major problems is that Oracle and Aurora have different ways of handling "savepoints". Although savepoint is an important DB tool for tracking or restoring individual transactions, an enormous amount of savepoints are created due to an extremely large number of orders that occurred on prime day, and the processing speed of the entire system abnormally decreases It is said that it is shown in the report that it has gone.

Mr. Matt Caesar, a computer scientist at the University of Illinois at Urbana-Champaign, looked at the material obtained by CNBC and said, "If Amazon was using Oracle, this problem did not happen," DB We point out that transferring the system is the biggest factor of obstacle. Furthermore, no prior measures were taken to match the size of the increasing transaction volume, and the preparation of coping manuals etc. at the time of problems occurred was insufficient, so it took 1 hour to remove the obstacle Points have also been pointed out.

By Nick Gray

In this way, Amazon's large-scale system failure seems to be undeniable as to the aspect that "happened to happen". Apart from this, Oracle Larry Ellison, co-founder and co-founder, has made a remark in the past saying that it is impossible without Oracle to say that Amazon has a policy of distance from Oracle I will. Also, Patrick Moorehead, an analyst at Moor Insights & Strategy, says, "AWS Aurora is designed for future-oriented applications, whereas Oracle is designed for legacy applications "I say.