A cloud by any other name

Today I received an announcement from sales about Jodo moving to a “cloud-based server infrastructure”. This is sales talking, so I just want to clarify exactly what they mean by “cloud”. :slight_smile:

From Wikipedia:
“The fundamental concept of cloud computing is that […] the processing (and the related data) is not in a specified, known or the same place(s). This is in opposition to where the processing takes place in one or more specific servers that are known.”

The announcement said: “All new accounts will be going to Windows 2008 based cloud servers starting now, there is a more detailed statement below the new server details.”

So, from the point of view of a reseller or shared user, what changes? My concept of cloud is that files are not restricted to a single server, and processing power is shared around as needed. So does this mean new account will have essentially 100% uptime, be unaffected by CPU spikes caused by errant users or a DDoS? Are files on one server being mirrored to another, so the crash of a single server in the cloud group doesn’t take the sites down on that server?

Whether my assumptions here are right or wrong, I’d like to get an idea how the “cloud” aspect affects the performance of these new accounts. Can this be explained a bit?

no cloud based still has instances that do restrict CPUs and ram etc, as windows instances require these to work :slight_smile:
but they are flexible in the manner that we can add on demand CPU and RAM if required.

Even in amazon, and many others you still have max limits that can hit in the instance if you have an OS there.

There is another sense of the ‘cloud’ in that there is no real ‘frontend’ server
For hsphere to work, we still have to have a front end. At some point it MAY be possible for us to make a load balanced frontend that allows multiple windows instances to reply requests, and we are certainly looking into that, but at this time they are cloud instances, running on elastic CPU, RAM, network, and storage infrastructures.

Essentially what makes a cloud is a storage medium that is stored on the network with large ‘nodes’ that work together, however an instance and only access limited numbers of these at one time, and in this case windows will max at 4 CPUs with the Hyper-V infrastructure we are using. We have looked into some of the VMWare a bit, but it would not be cost effective for us to move, and in all ways still limits the windows instances as this is
vSphere up till Enterprise edition limits to 4 CPUs, just as Hyper-V does, and enterprise edition Vsphere allows 8 vCPUs for windows to address, so you still have some level of ‘maximum’ per instance.

however with this we can do things differently.

Let me use CF for example. ODBC is notoriously bad on it, taking down everything.

We are about to change all that in how it works. We are going to make many small instances, with less RAM and CPU allocations than the dual cpus, quad cores, dual cores etc we use now, putting instead just a dual vCPU, 2GB RAM instance, with only a handful of accounts on it, this will in turn not SOLVE the odbc issue, but it will prevent it from being noticed as widely since there will be less accounts per ‘server’ that Hsphere sees.

BTW there is a lot about what exactly a ‘cloud’ is, and we hope to address all those in the coming months and years, not in definitions but in functionality and services, no it won’t all be on Hsphere as that simply won’t support it.

What Mosso did with allowing php on linux instances asp.net on windows instances in a near seamless manner wasn’t and isn’t as rackspace cloud a perfect solution, but it was a nice step adding the load balancer a level of intelligent routing by platform as well as availability. I am not saying we will emulate this in any way, but it was a nice thing that was removing the ‘os’ from the user equation, which will in the end be the ultimate ‘cloud’ solution as I’d vision it. there will always be limits, a cloud doesn’t magically remove limits at all. in fact it introduces some, what we’ve done here doesn’t require special coding etc. To work with these past solutions sessions had to be re-written, using DB backends only, or various other session passing workarounds, making a lot of complexity. What we’ve done is make the instances on top of it, integrated to hsphere for the best of what we can offer to the user right now, without requiring extensive changes in billing, setup or plans. In fact there will be some plan updates soon :wink:

Thanks Stephen, that clears up much of what I was wondering. It will be interesting to see where this goes in the hosting area, but as you say it depends on how Windows and HSphere evolve.

Rackspace cloud says merry christmas and !!!ERROR!!!

…Today December 23rd 2010, at approximately 11:02:35 UTC, your host server became unresponsive and after extensive troubleshooting and attempts to bring it back online, we determined that it experienced a catastrophic failure. All customer data on that host has been lost. If you have backups, we strongly urge you to create a new server from that backup, and we will share the old server’s IP address over to the new one…

Rackspace Cloud Server Outage | Web Hosting Talk

i said something ‘good’ about it, and it fails, ouch.

heh.. wow, that sounds pretty serious.

See it’s one thing to say “cloud” because things like CPU or disk space is “scalable”, but if files aren’t being mirrored across at least a couple of servers, then you’re down to a single point of failure, which kind of undermines the whole idea.

So I’ll go ahead and answer it since some will ask, yes we’ve found a bit of a dent in the armor. The nodes have a network level weakness as we’ve seen today with our own server(aka instance), the backend networks were all up, but front end down.

However the jodohost and vzsql nodes are not (right now, will be in next 15 hours) running on the same hardware as the other servers we’ve brought live and didn’t have server grade NICs like the nodes we are using for others.

So while any nic can fail at some point (and we’ve certainly seen that happen in Miami on a couple of servers), it should be a very rare event overall. I don’t think we’ll be able to make a fully redundant public interface immediately for teamed/failover public NICs, but it will be a consideration and option as we go forward in the future with enhanced offerings.
We think long term that instance/service level failover will be a better investment, and option, than a public nic teaming on the public interfaces.