Slashdot Mirror
Google Apps Gets a 99.9% Guarantee
David Gerard passes along a posting on Google's official blog announcing that they have extended the three-nines SLA for the Premier Edition of Google Apps from Gmail alone to also cover the Calendar, Docs, Sites, and Google Talk services. 99.9% uptime translates to 45 minutes a month of downtime, and the blog post puts this in context with Gmail's historical reliability, which has been between three and four times as good over the last year (10-15 min./mo.). It also claims, based on research by an outside group, that Gmail's historical reliability beats that of in-house hosted solutions such as Groupwise and Exchange, on average. Reader Ian Lamont adds an article in The Standard that digs down into the details of the SLA, revealing for instance that outages of less than 10 minutes aren't counted against the monthly 45 minutes.
It's called a cluster, "The cloud" is a really annoying buzzword for software as a service.
/Mikael
Greylisting is to SMTP as NAT is to IPv4
The issue is your internet connection AND your ISPs connection to the world. Your connection to the world is more likely to go down before a Google cluster would. Think of how often Telco's, ISP, and major hubs go down. This is the point behind having LOCAL copies of apps/servers/services, the odds that the hub/switch dies (with nothing else inhouse to patch around) is very slim compared to the odds of internet connectivity going south.
"The price good men pay for indifference to public affairs is to be ruled by evil men." ~Plato (427-347 BC)
The service they sell isn't beta. The service they give away is what they inflict new features on.
http://rocknerd.co.uk
I achieved four nines (%99.99) 8 years ago with Netscape's broken mail server "Suite Spot" running on a (at the time) three year old Sun E450 with 4 gigs of RAM. As I recall, it served about 120,000 clients on a large cable network in Chicago.
This whole "new web" thing is very pretty, but it seems like about three steps back to me.
It's a King Arthur cloud, maaan. Get with the times!
You can hold down the "B" button for continuous firing.
There'd be no need for a Beowulf-type cluster in this case.
Have a bunch of machines running identical instances of Apache, and randomly fire requests at them individually. This balances the load, and ensures that the servers themselves aren't a single point of failure.
It's quite a bit more complicated than this in reality, although you should get the basic idea.
Beowulf is typically used for clusters that seek to emulate a supercomputer (usually for scientific number-crunching), rather than a server. For this reason, something like Google's setup would more typically be referred to as a "server farm"
-- If you try to fail and succeed, which have you done? - Uli's moose
Google doesn't have 100% uptime? They have never gone down when I've noticed, guess its that sweet cloud setup they have there.
Seriously? I see it happen at least once every few weeks or so. It's usually very temporary, like as in less than a minute, but I'm quite familiar with the look of Google's error/service unavailable page...
"Convictions are more dangerous enemies of truth than lies."
The concept of "unplanned downtime" seems to originate in the banking world, where something as benign as daylight savings time could force you to take down the mainframe for two hours. It has unfortunately spread to other industries (healthcare records management pops up). The real question is if Google's application architecture requires planned downtime for the service as a whole or individual users.
Based on their roots, I would expect them to be able to do any upgrades in the ten minute window they exclude from their SLA.
Google appliance unfortunately is just for search. Here's to hoping they add app support as well in the future.
If you need to search through your 100GB of indexed documents, you want to be able to transparently break up that search query over multiple machines.
Actually, it's building the index of the documents that is especially computationally intensive. Particularly chunky is the algorithm to assign a significance score to each document. Once you've done that, actual searching can then be done by merging streams of information suitably, which it is pretty easy to do fast.
"Little does he know, but there is no 'I' in 'Idiot'!"
i had a 2 day outage which just said 'sorry an error has occurred' there is no way i could be without professional mail or documents for 2 days. I would only consider this for social non-critical uses.