The cloud of unreliability

It's not clear why anyone should be surprised that Gmail, Amazon.com's cloud services, Salesforce.com, MobileMe, or Netflix have periods of instability or downtime. These services are not promising five-nines of uptime, and they are dependent on complex code and a vast network "tubes," as the beleaguered Sen. Ted Stevens has said, to deliver bits to users. Services such as Twitter have set a new standard for unreliability, making the other cloud-based services look good in comparison despite their outages.

The much-ballyhooed cloud from which Web services emanate is inherently unstable and prone to odd behavior from any number of causes. At the same time, the Internet overall is incredibly robust and redundant. You just don't want to be caught at the intersection of some errant configuration change or badly behaving router. In the case of a Gmail outage, you need to have alternative e-mail services that capture messages from multiple sources to stay afloat.

Over time, the complex network systems underlying the Internet will become more reliable, but don't count on the Internet of 2008 or even 2015 to be operationally flawless. If you are not careful and proactive, the cloud will rain on you without warning.

[dburr13]

I agree...The cloud will not be ready for prime time for a long time...

Also i'm not sure the world is ready to trust any outside entity with that much power and control over thier data...I really believe control is a big issue here...The companies that create and market the programs that individuals and corporate customers use are wanting to build a system in which users are more dependent on them...I don't trust the big guys enough to believe that they will not attempt to exploit this control to extort higher and higher fees for the use of this service once it becomes embedded in day to day life.

[Mr. Dee]

I guess this is the part where Microsoft says: 'I told you so'. Microsoft has it right Software + Services. The local experience is still needed and with the recent outages from Apple and Google I am confidently saying that Microsoft will dominate this field. All this momentum and over enthusiasm for cloud services is still a bit too early.

[Penguinisto]

@ Dee: MSFT has experienced its own share of outages... and it also preaches all about "the cloud" long and loud.

Also, considering that security (or lacks thereof) are of equal concern "in the cloud", MSFT's track record has been at the near-bottom of the heap.

There are no exceptions to what is essentially a global rule: sites go down occasionally.

[Penguinisto]

@ Dee (again): This is a (partial) list of MSFT outages:

Hotmail (2008): http://www.internetnews.com/infra/article.php/3730541

Xbox Live (2008): http://www.gamespot.com/news/6184323.html

MSN Messenger (2005): http://www.msnbc.msn.com/id/6936783/

Windows WGA (2007): http://news.zdnet.co.uk/itmanagement/0,1000000308,39288927,00.htm

Microsoft Network (2001): http://news.cnet.com/2009-1001-251651.html?legacy=cnet

[amigosito]

Smart hosting service providers are turning to VMware to overcome this problem. Pardon the Kool-aid dispensation, but the VMware hypervisor has been clocked at about 3 years of uptime, and their enterprise suite includes the ability to automatically restart virtual machines if Windows goes down, or to failover entire datacenters worth of VMs to another location in the event of sitewide outage. It's pretty cool stuff.

[imddude]

You're on the highway, cruising 200km/hour. Your engine stops. It happens over and over again. You complain. Car manufacturers starting building in devices to monitor and restart your engine each time it fails. You think it's pretty cool.

[Penguinisto]

As someone who uses VI and ESX, err, no, VMWare is not a one-size-fits-all solution. Not even close.

Besides, I have yet to see a hypervisor replace a Cisco router. Network outages aren't just caused by goofs at the server level - they're just as often caused by faulty or mis-configured routers. They can just as easily go down due to an overeager backhoe operator.

[DarkHawke]

Amen, brother! Trusting even your personal computing, let alone business applications, to "the cloud" is like expecting the government to take care of all your needs: expect it to be slow, complex, cumbersome and A LOT more expensive than it seems. If my home system goes down, I pretty much know at whom to point the finger. Trying to nail down the culprit in a "cloud computing" failure scenario is about as easy as trying to lay hands on an actual could!

[imddude]

Yes, the government sucks. It created the Internet for you.

[Penguinisto]

@imddude: Incorrect. The US government invented the Internet for it's own use. Academia and popularity got it into the rest of the world's grasp.

[imddude]

The US government indeed claims it created the Internet for it's own use. You may believe them, I don't. The US government until today donates billions of dollars per year to high tech companies (direct investments in IBM, SUN, ...), supposedly "for it's own use". I found it rather illuminating a couple of years ago Bill Gates was crying like a baby when he heard government was reallocating subsidies meant to go to his industry to other high tech sectors.
In any case it's interesting to note that without the government the ICT boom of the late nineties never would have been possible.
Even modern IT management techniques such as ITIL are straight copy/paste from the Britisch bureaucracy.

[imddude]

Q: Why do managers like cloud computing?
A: Because they are with their head in the clouds all day anyway.

Q: Why do managers like virtualization?
A: Living in virtual reality all day anyway, it feels like a homecoming.

[djc8080]

Dan,

You've written about outages. Day to day, Yahoo Mail puts customer email in the spam bucket; Google Analytics misses 10% to 30% of the page views; and remote Facebook features simply don't work. Soon, Google will have a cookie nightmare with the integration effort of their hundreds of projects. Web companies can execute better, but the complexity of hundreds of client/OS versions multiplied against hundreds of server/OS combinations creates havoc.

The Internet has a redundant, fail-safe design that allows each node to compensate for failed nodes. TCP retransmits using any available route when packets fail to arrive as expected.

Businesses often have a centralist, integrated view of their web presence. Cloud services can be more affordable (i.e. free), but the single method approach is dangerous. For my business email in the Cloud, we use multiple services - never depending on one service. That includes the buggy desktop as one of the optional methods. For CMS, it's best for businesses to mash-up services across many free services like Blogger, Wordsmith, Facebook, and Myspace. Four services monitor our web projects (soon to be reduced to two.) This strategy avoids 100% fail for the business when one service dies or fails to deliver.

ROI and outsourcing trends dictate cloud computing as the future. Users should understand the need for redundant, distributed strategies for their applications.