Average EC2 Uptime?

Question

Curious as to 99.95% uptime REALLY means; Is it really going to go down 7 minutes a month? Please post your longest/average uptimes on EC2, thanks.

Answer 1

I've never had downtime on EC2, however, I do keep local backups and make daily images of my machines and port them to another availability zone, just in case. I use twilio to alert me if a machine is unreachable with a phone call to all my devices. Then I can just log in to EC2 and fire up a machine in another availability zone; worst case I'll be down for a few minutes.

Which in my case, is potentially pretty sucky, because my machines are doing 24/7 Forex trading.

My rule: know the potential cost of downtime, and be willing to invest that much in redundancy assuming it will happen - because it will.

That said, EC2 has never let me down. Helps probably that my servers are not in an area of the country where natural disasters are common. If you're in an earthquake zone, tornado alley, or a potential hurricane path, downtime truly is an inevitability.

Answer 2

SLA... Guaranteed uptime...

These are all very nice taglines. But when the servers aren't available for an hour (March 1, 2012, in the EU region) and the clients start calling, then it won't help you that they still have a 300 days uptime.

And when the lightning struck 1 out of 3 of their datacenters in the EU, we all found out that they have no off-site redundancies, and the fact that they have 3 datacenters doesn't mean a thing.

One must love the phrase "degraded performance", that actually means: "cross your fingers and pray that your data will still be available after the catastrophe passes".

I'm still trying to look for any official/non-official statistics about the availability percentages of all of their datacenters.

No luck thus far...

Answer 3

SLA's are useless. They only measure how much risk the company is willing to take on and have no bearing on actual uptime. I've seen SLA's, with heavy penalties, offered when the company knew the could not meet the SLA in order to land the sale.

I have one client with 400+ days of EC2 uptime and another with 300+ days as measured by web pulse, this is by far the most reliable service I've worked with.

Answer 4

For my single instance running in the US-East availability zone, 9 months, 0 downtime.

Answer 5

Read the SLA carefully (http://aws.amazon.com/ec2-sla/) they only count "Region Unavailable" as downtime, and what is more they only count it as downtime if the region is down for 5 consecutive minutes.

“"Annual Uptime Percentage” is calculated by subtracting from 100% the percentage of 5 minute periods during the Service Year in which Amazon EC2 was in the state of “Region Unavailable.”

By my count this mean any downtime of less then 4 minutes is not countable. Also if they do break the SLA they are only in for %10 of the month in which you had largest downtime bill. So if they where down for all of January and your bill was $100 they would apply a $10 credit to your account.

I would have a hard time convensing my boss that this is a serious product with a SLA like that.

Answer 6

conrad@papa ~ $ uptime
 04:42:36 up 495 days,  8:51,  8 users,  load average: 0.02, 0.02, 0.00

Answer 7

Checking out the AWS Service Health Dashboard will get you a good idea of any current or past issues. My experience is that the AWS uptime is better than most "traditional" hosting options (even full-blown redundant RackSpace setup...).

However, simply going with AWS for uptime is like getting a car for the keychain (ok, almost.. ;)). With an architecture utilizing AWS the big benefit is scaling (without upfront costs).

Answer 8

Since Amazon switched to provide an SLA, I've never had an instance go down on me. When I've had instances go down in the past, Amazon has always sent a message informing me that the instance is degraded before it actually disappeared, so I've had time to start up a new instance.

The previous answer makes a good point, though; EC2's service model dictates that you write your apps to handle failover to a new server if you're not prepared for extended down time.

Answer 9

Usually uptime is calculated in a yearly basis. So if you have a Service Level Agreement for 99.95% this means:

365 * 0.0005 = 0.1825 days or 4.38 hours

If during a year of service there is an outage and your system is down for more than that, then you are liable for compensation.

As of your question, I have a server running unstopped in EC2 for about 3 months now. I would say that their uptime is good, but if you have a mission critical application you definitely need to have a fail-over solution. A good uptime only means that they will be able to respond to an outage quickly. Even a 99.9999% uptime won't be able to save you if you aren't prepared for an outage.