Slashdot Mirror
Amazon's Cloud May Provision 50,000 VMs a Day
Dan Jones writes "It has been estimated that Amazon Web Services is provisioning some 50,000 EC2 server instances per day, or more than 18 million per year. But that may not be entirely accurate. A single Amazon Machine Image (the virtual machine) may be launched multiple times as an EC2 instance, thereby indicating that the true number of individual Amazon servers may be lower, perhaps much lower, than 50,000 per day. So, even if it's out by a factor of 10 that's still 1.8 million VMs per year. Is that sustainable? By way of comparison, In February of this year, Amazon announced S3 contained 40 billion objects. By August, the number was 64 billion objects. This indicates a growth of 4 billion S3 objects per month, giving a daily growth total of about 133 million new S3 objects per day. How big can the cloud get before it starts to rain?"
who cares how many potential VMs the "cloud" can host. its methodone for most end users/devs real problems: inefficient code. the "just pitch machines at it until it runs fast!" mentality will catch up to us.
That's not true. We use Amazon's cloud to host some of our servers. The reason we do it is for two main reasons. (1) We don't need to worry about equipment maintenance. Let me repeat that lest you think its not a big deal: We don't need to worry about equipment maintenance! (That is a big deal when you leave your basement but don't necessarily have a dedicated IT staff). (2) We are in a rapid growth phase. We cannot estimate well enough what are computing needs, our storage needs, are going to be 1- 2- 6- months down the road. We also don't have $50k to drop on equipment and storage that may be utilized 6 months from now, but we sure as hell know if we bought it now it wouldn't be used immediately. Amazon's cloud makes it trivial to keep up with our growing demand without paying up front for it. Sure we pay more to "rent" the stuff from Amazon, but its simply the big(O) argument: Amazon's pricing scales worse than the classic alternatives, but the constants out front are tiny.
My company tried to provision 10,000 amazon instances to perform scalability testing of our software that runs on many computers. The math was simple - 10,000 servers * $0.15 / hour = $1,500 / hour for testing. We liked the multiple OSes & versions (Linux - Redhat, SLES, Windows - 2000, 2003, 2008?) and software stacks (mysql, apache, websphere, sql server, iis, etc...) that we all available out of the box.
However, if you need more than 20 servers, you have to fill out a form. A sales rep and tech guy called to discuss our needs. It turns out that they could only handle around 1000 instance request across all data centers unless we "reserve" the machines at $300 / each, which blew the math - 10,000 servers * $300 = $3,000,000 to start.
Looking at the article, it is likely that people are re-requesting the same machine be started & stopped multiple times per day - 50,000 is probably off by an order of 10.
I thinl you're missing my point - why have multiple OSes if they're all the same type of OS and the apps could all happily run on the same OS instance? As for deployment - have you never heard of a tarball? OS dies - take app tarball to new server , untar. Hows that different to copying a VM machine file over?
In the real world, people run apps like Exchange or Oracle, which take hours to install to a vanilla state, and that's not counting the potentially terabytes of data associated with them.
Even the most primitive "tar ball" Linux app will have dependencies on the OS, and those can and will eventually break, unless you freeze your OS version forever. If you have enough apps and servers, that will become a nightmare to manage. Do I upgrade or not upgrade? Will this patch or that patch break one of the apps? This is how people end up running Linux 2.2, or 32-bit Windows on 64-bit platforms, because migrating 1 app is hard enough, but migrating a server with 20 apps on it is a recipe for disaster.
Virtualization lets you quite literally drag & drop a running host OS from server to server. During maintenance time, that's like magic. No more 3am hardware replacement jobs for me! You can clone a machine while it's running, isolate the clone onto a virtual network, and test an upgrade without interrupting users. Sure, you can do that with most backup & restore tools, but VM platforms do it quicker, and with fewer admin steps. You don't even need spare hardware.
I once replaced every single hardware component of a running VM farm, servers, cables, switches, even the SAN, while it was running. During the day. Zero outage, no packets lost, no TCP/IP connections closed or user sessions disconnected. We even had terminal server (Citrix) and console (SSH) users on. Not one user even noticed what was going on. I'd love to see you try that with 'tar'.