Hacker News Reader

Comment by panarky | Hacker News Reader

Bigger, faster, more open, more secure ... what's missing from this list?

Oh, that's right, how about more uptime?

  Provider               Outages  Total Downtime
  ---------------------- -------- --------------
  Amazon EC2                  12     2.01 hours
  Google Compute Engine       65     3.27 hours
  Microsoft Azure VMs        102    42.80 hours

Source: https://cloudharmony.com/status-1year-of-compute-group-by-re...

And this doesn't show the other Microsoft cloud services that have even worse reliability.

At my company we invested a lot of time trying to make Azure Service Bus and Visual Studio Online work. Both were so dreadfully unreliable as to be unusable.

View on HN

For IAAS providers, you should care about the uptime of the regions (datacenter) that you are planning to use and the uptime of the specific services that your vms will use within those datacenters. In particular, I recommend paying close attention to the monthly/quarterly uptime trend in addition to aggregate uptime numbers.

And if you really care, try and find out the datacenter(s) your providers will be using and find out the historic performance of their different infrastructure and redundancy providers, and how often they actually test and upgrade their systems.

Look at the difference now when I filter by region and add the region data to the chart... it clearly shows just how hard these comparisons can be:

  Provider               Uptime   Outages  Regions Total Downtime
  ---------------------- -------- ------- -------- --------------
  Google Compute Engine  100%        0     1         0 hours
  Amazon EC2             99.9984%    7     3         0.13 hours
  Microsoft Azure VMs    99.942%     56    6         2.89 hours

https://cloudharmony.com/status-1year-of-compute-in-america_...

Yeah, adding downtime between regions seems like it would mess with the data. Like if you have a single region with 40 hours of downtime, that's much worse than 10 regions with 4 hours of downtime. But represented as the same total.

Hey, at least it's not Veriozon cloud which just told users to power down everything today in preparation for a scheduled 48 hour outage.

http://www.computerworld.com/article/2865802/verizon-warns-e...

In CloudHarmony's assessment of 49 cloud compute providers, Microsoft Azure was literally dead last in total downtime.

Number 49 out of 49 services.

If Verizon is offline for the entire time, maybe they'll be number 50.

Wow. That’s just bizarre.

Can't possibly be accurate. It's missing the major AWS outage from last year that took down the entire US-East zone for more than half a day.

US-East is a region made of multiple availability zones. Not all of those zones went down (none of my stuff was offline in 3 of the AZ's, for example).

You disingenuously left out the Region column from your copy of the table from the article you linked, which entirely alters the meaning of "Total Downtime". Any explanation for that?

That said, I'm a GCP certified engineer. I love GCP, but I can see Google losing the cloud war to Microsoft solely in that Microsoft has a more comprehensible offering (UI and libraries/services) and, most importantly, knows how to market its platform.

AWS is more comprehensive than both by quite a wide margin. MS does indeed have a long history of relying very heavily on marketing. Time will tell if marketing can paper over the large gap in features.

More comprehensive by a wide margin? Can you unpack that? I get the impression that there is no functional equivalent to Google App Engine on the Amazon side. You still have to "reserve" dynos, you still have to define how your app scales. There is no "autoscaling" feature that's like App Engine on the Amazon side. Or at least if it exists, it's not as mature as App Engine.

Is this the certification you are talking about :

https://cloud.google.com/developers/training/exams/

Those are the pre-requisite for the 400-series exams. I had to pass all those just to get into the 400-series classes. Then you have to pass the exams and then do a project for each part of the platform. I did all that.

Could you please link me where I can find more information about the 400-series exams / how to become a certified GCP engineer?

I'm not really sure! I was invited into the program through the partner channel at Google because I was working for a company that had already done a lot of development on the app engine platform. I was one of the early adopters of GAE, as I started using it the first year it came out. One of the apps I co-wrote became a case study Google used.

See this is kind of an example of how Google is awesome at engineering but short on communication. I can't even explain how to get into their fold, so to speak.

There was a pretty painful, several-hour Azure outage recently, but 102 outages is over an order of magnitude higher than my Pingdom/Clicky uptime data for a few sites in US East last year.

I have a single instance and other than that big one, my monitor hasn't really gone off, other than during patching. Im not sure what those hours are referring to.

Probably includes partial outages.

I'm going to show the Azure downtime numbers to my boss next to the downtime numbers for our private cloud next time I want to ask for a raise....

Way to set the bar...

When a company trumpets some bold statements for marketing purposes, I immediately think about the worst or most far-fetched interpretation of their language. I wonder if their marketing people are being coy and the honest truth is, in some ways, the opposite or far less than what they are saying.

With 'more' terminology, I always wonder what the original was. "More" than 1% secure is... not terribly specific. "Best ever" doesn't really mean it's good, just that it's always been worse.

As a side note, I'm not implicating Azure here, just a general context.

Maybe an automated corporate marketing speak to normal human language translator with UI like google translate could be cooked up?

You probably know that security includes availability. Just saying...

Reply on news.ycombinator.com