Ama­zon cloud stor­age out­age stalls in­ter­net

Ama­zon Web Ser­vices’ main stor­age sys­tem goes down for 4 hours

USA TODAY US Edition - - FRONT PAGE - El­iz­a­beth Weise @eweise USA TO­DAY

Four-hour hic­cup dis­rupts busi­ness

It didn’t quite break the In­ter­net, but a fourhour out­age at Ama­zon’s AWS cloud-com­put­ing divi­sion caused headaches for hun­dreds of thou­sands of web­sites across the United States on Tues­day.

Lit­tle-known to con­sumers fa­mil­iar with Ama­zon’s on­line shop­ping site, Ama­zon Web Ser­vices is a gi­ant provider of the back-end of the In­ter­net. For sites such as Net­flix, Spo­tify, Pin­ter­est and Buz­zFeed, as well as tens of thou­sands of smaller sites, it pro­vides cloud-based stor­age and Web ser­vices for com­pa­nies so they don’t have to build their own server farms, al­low­ing them to rapidly de­ploy com­put­ing power without hav­ing to in­vest in in­fra­struc­ture.

For ex­am­ple, a busi­ness might store its video or images or data­bases on an AWS server and ac­cess it via the In­ter­net.

While not all were af­fected by the out­age at one of AWS’ main stor­age sys­tems, some ex­pe­ri­enced slow­downs af­ter a big por- tion of Ama­zon Web Ser­vices’ Ama­zon S3 sys­tem went off­line Tues­day af­ter­noon.

The ser­vice is used by 148,213 sites ac­cord­ing to mar­ket re­search firm Sim­i­larTech. The out­age ap­peared to have be­gun around 12:35 p.m. ET, ac­cord­ing to Catch­point Sys­tems, a dig­i­tal ex­pe­ri­ence mon­i­tor­ing com­pany. Op­er­a­tions were fully re­cov­ered by 4:49 p.m. ET, Ama­zon said. The Seat­tle-based com­pany did not com­ment on the cause of the out­age.

The sys­tem that went down was the first of what now are three AWS re­gions in the U.S.

It is still the largest and is also where AWS rolls out new fea­tures, “so it’s dis­pro­por­tion­ately big,” said Ly­dia Leong, a cloud an­a­lyst with Gart­ner.

AWS be­gan as a prof­itable side­line to Ama­zon’s main on­line sales busi­ness but has since

grown to be­come the ma­jor player in the arena as well as a ma­jor money-maker in its own right for Ama­zon.

In the fourth quar­ter of 2016, the divi­sion ac­counted for 8% of Ama­zon’s to­tal rev­enue.

“This is a pretty big out­age,” said Dave Bar­to­letti, a cloud an­a­lyst with For­rester. “AWS had not had a lot of out­ages, and when they hap­pen, they’re fa­mous. Peo­ple still talk about the one in Septem­ber of 2015 that lasted five hours,” he said.

S3 has “north of 3 to 4 tril­lion pieces of data stored in it,” Bar­to­letti said.

AWS S3 is used by busi­nesses both large and small. “More than any­thing else, S3 cus­tomers need to be able to get at their data, be­cause of­ten S3 is used to store images. So no S3, no nice picture or fancy logo on your web­site,” Leong said.

That was ex­actly the prob­lem faced by Lewis Bam­boo, a small, fam­ily-owned bam­boo nurs­ery in Oak­man, Ala. “As our busi­ness is in bam­boo plants, pic­tures are a very im­por­tant part of sell­ing our prod­uct on­line. We use Ama­zon S3 to store and dis­trib­ute our web­site images. When Ama­zon’s servers went down, so did the ma­jor­ity of our web­site,” said the com­pany’s chief tech­nol­ogy of­fi­cer Daniel Mul­laly.

“Thank­fully we also store the images lo­cally, and I was able to serve the images di­rectly from our server in­stead,” he said.

The ef­fects of the out­age var­ied de­pend­ing on the site and how it used AWS. Mod­ern web­sites usu­ally pull data from mul­ti­ple data­bases in the cloud that can be stored all over the world, so a photo might come from one place, a price list from an­other and a cus­tomer data­base from a third.

For that rea­son, en­tire web­sites rarely go down, but var­i­ous parts of them may take a long time to load or not load at all, leav­ing bro­ken links or images.

Com­pa­nies have been steadily mov­ing stor­age to the cloud be- cause it is cheaper, eas­ily ac­ces­si­ble and more re­silient. But the down­side is that when there are prob­lems, there’s a cas­cade ef­fect.

“There are lots of peo­ple hav­ing a not very good day at the mo­ment,” Leong said on Tues­day af­ter­noon.

It’s pos­si­ble to con­tract with mul­ti­ple com­pa­nies to avoid po­ten­tial prob­lems, but it’s pricey, so many com­pa­nies make peace with the knowl­edge that on rare oc­ca­sions they’re go­ing to have that very bad day.

“Only the most para­noid, and very large com­pa­nies, dis­trib­ute their files across not just AWS but also Mi­crosoft and Google, and repli­cate them ge­o­graph­i­cally across re­gions — but that’s very, very ex­pen­sive,” she said.

Ama­zon wasn’t able to up­date its own ser­vice health dash­board for the first two hours of the out­age be­cause the dash­board it­self was hosted on AWS.

The most com­mon causes of this type of out­age are soft­ware re­lated, Le­ung said. “Ei­ther a bug in the code or hu­man er­ror.”


Newspapers in English

Newspapers from USA

© PressReader. All rights reserved.