Suppliers of cloud solutions: Everything fails. Prepare in advance

Power outage in a data processing center of Amazon and falling of services EC2 (Elastic Compute Cloud) put in the forefront shortcomings of the failsafe systems, but Amazon not the only supplier of "clouds" who cannot guarantee 100% of available time.

GoGrid, offering calculations within technology infrastructure - as - service (IaaS), implements a credit line when the operability of service falls lower than 100% for clients. But it does not mean at all that the cloud service never stops.

"Speaking about elements of service, we expect them readiness for 100%, but if it does not occur, we are ready to compensate it", - the CEO and the founder of GoGrid John Keagy says. – Everything fails. Clients should understand: the agreement about 100% the level of service is not 100% a guarantee of this level".

Clients will be able to hold applications in work during a downtime of DPC if are ready to work and develop the plan of recovery after accident. The clients of Amazon who do not have the plan and ensuring fault tolerance suffered from idle times last week, unlike those who planned something similar beforehand.

"All this is infrastructure", - told Kigi. - It is not the platform as service (PaaS) or the software as service (SaaS). It is the crude infrastructure requiring that the user bore a certain responsibility for how all this is implemented".

Amazon allows to place applications in several "zones of availability" for the mezzanine board, however is not clear how far there are from each other these zones. Last week failures affected several such zones.

When Foursquare, Reddit, Quora, Hootsuite and others were disconnected, success of the website photoexchanger SmugMug shows - as advance planning can help clients to endure that the CEO of SmugMug Don MacAskill referred to as Amazonapokalipsis.

SmugMug was spread on three zones of availability, but decided not to use Elastic Block Storage Amazon service because of "the unpredictable performance and fragmentary reliability", Makaskill in the blog wrote.

If you place crucial applications in "cloud", Makaskill advises, place them in different zones of Amazon (The western coast and East coast, for example) or with several suppliers of "cloud".

The service of balancing of loadings Amazon does not work in different zones so clients should make independently some extra work and use the software of third-party developers to secure themselves, the analyst of Gartner Drue Reeves says. Placement of applications at several suppliers of "cloud", is not something impossible, but it is difficult due to the lack of standards and interoperational incompatibility.

Josh Odom heading product development for a cloud platform of Rackspace company notes that application launch in several DPCs is the best method to guarantee 100% of time of smooth operation. He considers that a big problem is not the application, but its data. According to him, any type of replication with relational database systems is quite bulky.

Microsoft announced "Windows Azure Traffic Manager" recently, saying that the application will allow "unroll the application between topologically separated by DPC, allowing to distribute workload between these DPCs round, on the basis of schemes of fault tolerance and balancing of performance of loading". Now this application is available only in technology community, in the preview mode, i.e. it is not ready for use by customers.

While Windows Azure Traffic Manager distributes traffic between several DPCs, SQL Azure Data Sync also in the beta, replicates "databases on several DPC to resist to loss of power supply, declared in Microsoft.

The App Engine service from Google can transfer applications and data from one DPC to another without data loss or idle time, in case of refusal DPC, the product manager of Google Greg D'alesandre says. Google does not report, data processing centers are how deleted from each other, but it is said: "a system is developed so that there is no uniform geographical point of failure".

Meanwhile Amazon is accused of lack of a clear explanation: what last week in fact went not so. Amazon referred to "a network incident" which "included repeated duplication" volumes of storage of large volume, having created, thus, deficit of capacity and having lost contact with virtual machines.

Thorsten von Eicken, the technical director and the founder of RightScale, supplier company of the services increasing functionality of Amazon services told that Amazon "receives F" (negative attitude. – a comment of TAdviser) for communication and the unsuccessful offer of the analysis of the prime causes of an event.