Google is claiming it is much more reliable than M$ when it comes to Gmail/Hotmail.
“In 2010, the company says, Gmail was available to business users and consumers 99.984 per cent of the time – about seven minutes of downtime per month – and there was no scheduled downtime for end users. “Weâ€™re particularly pleased with this level of reliability since it was accomplished without any planned downtime while launching 30 new features and adding 10s of millions of active users,” the post reads.”
Presumably M$ uses that other OS do to its business. This is what I would expect knowing about that other OS’s re-re-reboots, critical updates and malware. If M$ cannot do any better than 46 times worse than Google, they will get their butt kicked out of the cloud in short order. They will have to rely on their sacred cash-cows, that other OS and Office, until people finally catch on to the fact that M$ doesn’t make good software.
A typical GNU/Linux server can do 99.9% up-time on its own. With an automated failover cluster you get Google-like reliability. You stay on the air as long as you can keep any part of the cluster up. The simplest way to get some of that reliability is to use DRBD (RAID 1 or TCP). This takes care of storage, hardware and power failures affecting one node at a time.
Suppose you re-boot once a month to maintain updated software on your server. If a re-boot takes 1 minute and there are no other failures, you have up-time of (86400 – 60)/86400 = .9999 . If you have two systems with similar up-time and you can control when the reboots take place, you have up-time of the combined cluster of 100% . Unfortunately, in the real world, accidents happen and 99.9% up-time for one node is about the best you can do considering all factors. With a 2-node failover cluster you could get 99.99% if no common disaster befalls both nodes.
From this analysis, I deduce M$ has a common disaster befalling all their nodes, that other OS. It is possible that M$ is so popular that it fails from overload, but that hardly is a defence for a business that’s “all-in” for the cloud. Netcraft’s Survey of the top 42 hosting sites shows only 4 in the top 42 run that other OS. Price and reliability probably move that trend.
In my own experience, using a single server with a backup server I get all the reliability I need for what I do. I do sleep sometimes and do not need services 24×7. Google does pretty well for cloud services using GNU/Linux. If M$ cannot do any better than a single server can, perhaps the world should not rely on M$ for IT. There is an alternative, GNU/Linux and Free Software.
Debian GNU/Linux has just entered “Deep Freeze” and could be releasing version 6.0 sometime this week. With 25000 packages on more than 12 architectures, having only 76 release-critical bugs is an amazing feat of reliablity. Imagine software that reliable on all your nodes. You can do it.