Storage Availability 🤔

Choosing the backup solution that fits within your budget and provides peace-of-mind is essential. As Imagen provides many options for redundancy, you can feel safe knowing your data is safe with us.

Single-point-of-failure 

Let's begin with a hypothetical thought experiment. Imagine a fire breaks out in the datacenter where your data is stored, and the next server on the rack is the one that started the blaze. Luckily, the fire has been extinguished.  Sadly, the whole rack has been destroyed in the process - including the primary server with all of your data.  

Scenario 1 - LRS

Fortunately, you've opted to use Locally Redundant Storage (LRS), so there is a duplicate copy of your data stored on another server.  Unfortunately, that other server is located on the bottom shelf of the same rack. Your data has been lost and cannot be recovered. 

Scenario 2 - ZRS

You use Zero Redundant Storage (ZRS), where a second copy of your data is stored - again, within the same datacenter.  This time, however, the duplicate is located on a server rack on the other side of the building. The fire was extinguished before reaching your data. 

Scenario 3 - GRS

You use Geographically Redundant Storage (GRS), so your backup copy is stored safely in a second datacenter, which is located hundreds of miles away. You may experience some disruption as the failover takes place, but your Service Level Agreement (SLA) will define the delay between disaster and recovery. 

Summary

Suffice to say the third option offers the greatest safety, but it is also often the most expensive. This balance of budget and disaster recovery is part of availability planning. 

It's also worth pointing out that the failover is, in itself, a single point of failure. Let's now review the nines (9s) that govern failover reliability.

The measure of availability

Availability is measured by a percentage of uptime per year.  For example, 'Seven Nines' (99.9999999%) would allow for no more than 3.16 seconds of system downtime per year. If your data were (as in the hypothetical fire scenario above) covered by a 'Seven Nines' SLA, the failover would have to take place within 3.16 seconds.

Conversely, one nine would allow for downtime of up to 36.53 days. 

Thus, the more nines you have, usually the higher the cost.