Guys,

I'm quite interesting in one benchmark for Linode vs Amazon EC2.
It seems to me that Amazon EC2 should be more stable (available) other than, say, Linode Fremont DC because it has numerous resources underneath: maybe it spreads its works across different DCs?

Is it true, or it is just my perception?

You can get an idea of the stability/availability of AWS on their status page here: http://status.aws.amazon.com/

I'm not sure what you mean by spreading across DCs. As far as EC2 is concerned, an instance is hosted in a specific data center (I think Amazon calls them availability zones). Same as with Linode. You can always move to different DC, or run a backup server in another location.

You can engineer resilient services with both platforms just fine. Some things are easier with EC2, other things are easier with Linode. However, there is no magic faerie dust that can turn a single instance anywhere into a resilient high-availability service, and EC2 provides no inherent magic multi-instance availability faerie dust, nor scaling faerie dust, that Linode doesn't also have.

(EBS and S3 can help out, but they're also failure surfaces with significant limitations. Again, no magic faerie dust.)

_________________

/* TODO: need to add signature to posts */

Amazon puts multiple availability zones in each datacenter. They made this clear during that big outage a while back.

What does it mean? Does it mean that Amazon EC2 is more stable than VPS?

What does it mean? It means something iml said was wrong and Guspaz was providing a correction. Also, "than VPS"? EC2 *is* a VPS service.

All it means is that when using Amazon Web Services, data center-wide issues such as power failures can affect more than one of what they term Availability Zones. This is something for their users to keep in mind.

This is not to mention that their recent outages also demonstrated something else everyone should already know: that in complicated systems, outages can have impacts even beyond what was expected. For example, some of the region's central management systems were overloaded by the errors caused when they tried to interact with the AZ that was (mostly) down.

The RFO is a good read.

_________________
Matt Nordhoff (aka Peng on IRC)

An "availability zone" is a datacenter. A "region" is multiple datacenters near each other.

An EC2 instance *is* a virtual server/VPS/whatever -- it's the exact same thing. It's a single VM instance running on a single host somewhere in a rack of machines somewhere in a datacenter. There's no magical difference other than marketing. Don't be fooled.

-Chris

Guys, you see, for me is quite important availability and I don't know why, but Linode website/community can't assure me that I will have it.

Do you have some stat data: is Linode more available other than Amazon EC2?

Guys, seriously, Foursquare and Quora are Amazon EC2 clients.
Foursquare has around 7.5M users per month
https://www.google.com/adplanner/planni ... 01&lp=true


We expect around 1 million users, as we are going to invest lots of money into online/offline ads.
We want availability. But what I hear on this forum, that nobody grantees it.
Does Linode have notable clients with 3mln+ users per month?

If you assure me, we will definitely subscribe to Linode 4096 plan.

Thanks.

OK, caker disputes what Guspaz/I said about the meaning of AWS's availability zones. From a couple minutes of Googling, it seems to be consistent to "data center", but Amazon is unwilling to actually use those 2 words. In any case, what I said about AWS occasionally experiencing region-wide outages is true.

Back on topic, TheClient, nobody really provides guarantees, and even when they do, such guarantees are of questionable value. The bottom line is, any decent company does the best they can, and mostly succeed, but nobody is perfect and everybody suffers occasional, unexpected outages. Edit: You might find how companies *respond* to outages more interesting. Amazon has a reputation for poor communication during outages, and IMO Linode could do better there too, but both of them post good RFOs afterwards. And that's not even getting into how well and quickly they resolve problems, and what measures they take to prevent them from recurring.

Hurricane Electric (Fremont)'s power situation aside, Linode and the data centers it uses have acceptable (i.e. excellent) track records. I imagine the same is true for AWS. They are both good platforms to build reliable services upon, but in both cases you have a responsibility to build the level of reliability you desire. For example, to work around a Linode or EC2 host failing, have two VPSes in an HA configuration. To work around a data center power outage, have VPSes in multiple data centers/availability zones. To work around single nuclear attacks, use multiple AWS regions (Linode's data centers are all far enough apart). To work around large-scale nuclear war, put them on different continents, though such an event would probably cause widespread power, Internet and life outages for your users, making your service's reliability irrelevant. To workaround Rackspace acquisitions, use both Linode and AWS at the same time.

Edit: As for large customers, I'm unable to provide a list, but a Linode employee might chime in. (Or another user who feels like digging up one of the lists previously given.)

Edit: My NTP server has hundreds of thousands of users, possibly a couple million over the course of a month. Do I count?

_________________
Matt Nordhoff (aka Peng on IRC)

Never believe anyone guaranteeing 101.21% uptime, SLA or whatever. If you need HA, then build a HA setup yourself.

With Linode you can chose datacenter location and then nicely disperse your points of failure geographically.

Some actual data from a single Linode 512 running in Newark: http://stats.pingdom.com/mfbtpxiytz5c/243361

Linode press releases have named Creative Commons and The Onion as Linode customers; I've personally verified the latter, but haven't investigated the former. There's others out there too that aren't named in press releases.

I don't think the Linode plan size has much to do with availability, nor with the level of support. What will make a difference availability-wise -- and this is something the Reddits and Netflixes and The Onions of the world do -- is eliminating single points of failure as much as possible. Even with that, the Reddits and Netflixes still fail for hours on end.

_________________

/* TODO: need to add signature to posts */

What they say:


What I think:

^this.

If you need HA, then build HA, don't buy a single server and think it's HA.

Caker is right, I was misreading Amazon's root cause analysis.

Also, agree with Azathoth, high availability setups are just as much the responsibility of the customer (deploying a fault-tolerant multi-datacenter infrastructure) as the provider.

During the big EC2 outage, Netflix lost a huge chunk of their infrastructure, but didn't suffer any outage, because they designed their infrastructure to tolerate chunks of it disappearing; they spun up more instances elsewhere to compensate.

Such setups are obviously more sophisticated than most people want, but keeping hot or cold standby servers that can be switched to very quickly gets you a lot of the way there.