Big clouds need a tightly integrated stack, and a small number of hardware components that can be upgraded and swapped out en masse. This is precisely what Verizon lacked in its first cloud, and what it is gaining with a new service. The new Verizon public cloud was announced by the company on Thursday, and will see it be cost …
Another day, another cloud provider..
Saying that, a presence in London makes it slightly more interesting, the only real major Cloud player I can think of with a UK data center is Rackspace. Aside from cost, its a primary factor why many people choose not to do it.
When your operating (or aiming to operate) at such a scale the traditional tools typically fall apart pretty quick. Because there's so few organizations operating at such scale, a lot of the problems to address are relatively new and there are only so many people in the world that have experienced them and have ideas on how to solve them. Thus the likes of Google, Amazon, MS, and now Verizon are/will be using a ton of custom in house stuff to manage their clouds because the traditional players just don't have the vision (probably because they don't have the customer base that needs it) to scale in that manor.
I do find it puzzling that Verizon chose an obsolete hypervisor like Xen over KVM though. With Amazon I can understand, their stuff is old and customized to the point where it's hard to make a change. But to deploy Xen in modern times just seems like a really poor decision.
You don't have to look very far than the bulk of your typical vendor "global management system" products to see how poor they are in general. None of it of course was ever intended to be hyper scale, so it falls apart.
Don't think for a minute though that building for this kind of scale makes sense for a SMB. I've worked with a couple ex-amazon folk who all they knew was the amazon way, and they worked to try to replicate that elsewhere - pretty much universally with catastrophic results. You need fairly radically different approaches for small(ish) scale vs absolutely massive hyperscale. Trying to shoehorn one of those to do both is just shooting yourself in the foot -- with one exception -- that you know -- with certainty that you will go to hyperscale fairly quickly. I'm not talking about some bullshit social media company that has aspirations to become the next facebook. I'm talking about something serious, and something closer to reality. Not likely to be an organization I will ever want to work for myself - I have more fun at smaller enterprise scales, where I can work across silos. I cringe at the thought of working for a big service provider.
For a good example of this you can see this page(sorry for the link) where I rip through the hype of SDN and really what a scam the networking industry is trying to pull on (most) of everyone. SDN can make sense for the Verizon's of the world, but it doesn't make sense for the bulk of customers out there:
(It literally took years before I figured out precisely what SDN was - got so many conflicting explanations - so I got the answer from the person who invented it(which confirmed my original suspicions) and I felt better about writing on it more directly).
Of course with this new design it sounds like the overall quality of the Terremark/Verizon cloud will go down, perhaps as low as the quality coming out of Amazon (never having used MS or Google clouds I suspect their quality is similarly pathetic), which is unfortunate, but their previous cost model was just about as unfortunate.
A few years ago Terremark proposed to the company I was at a virtualized data center for disaster recovery which had a $3 million installation fee along with ~$150k/mo in monthly fees - this was running on top of their (at the time new) Cisco UCS infrastructure.
I could of bought equivalent HP gear (and 3PAR storage before HP bought em) and used it in house for around $700k and about $20k/mo in hosting (all tier 1 stuff with 4 hour on site support etc, vsphere enterprise plus, no shitty whitebox cutting corner stuff), it was not even two full racks of gear.
Terremark didn't understand how easy it was to manage such a setup("do you have the staff to be able to manage that?" -- me: "dude! it's 12 heavily loaded blade servers, a few switches, a storage array, a NAS gateway, and a pair of load balancers!! how hard is that really!?"), as much as my management could understand Terremark's pricing.
To Terremark's credit though they were the only cloud provider to have the balls to quote us actual numbers, the other players bowed out before spending too much time (one of them gave me a cost of $1 million just so we could stop wasting everyone's time).
They gave us another number, the first number - which had only a $10 setup fee -- though the monthly cost for that was $272,000/mo (not on Cisco UCS). So, obviously a pretty quick ROI for in house.
Terremark is not alone with numbers like this, I hear similar stories from other folks at other providers. Even Amazon's IaaS pricing structure is terrible(mainly due to 90s era architecture which leads to a large amount of wasted resources).
Re: makes sense
That was longer than the article.
Re: makes sense
You are quite correct (nice write up, btw) that SDN doesn't solve your problem, but then it's not intended to. SDN does not do anything to make traditional N-tier apps faster/sexier etc. despite what breathless hype might say. What it does (or, more cynically, what it promises to) do is allow you to build and control the network services that your app needs more effectively. SDN for me is the ability to deliver network services as a virtualized function. Instead of specifying "I'd like a Cisco XYZ firewall" you request services and functions and the network controller delivers them "I'd like a packet fitering service over here please".
Let's say that Mega Telco Inc's product managers wake up one day and decide to implement Hosted IP PBX. In a traditional telco this is a PITA. You have to go off and provision some app servers, a (SIP) switching core, SBCs, customer IP phones -- all horrible. So you get through that then your CEO says, "expand to Canada! Support encryption! Go faster in Arizona!" Now you have to do the whole thing again and work through some regulatory stuff too. No matter how good at Broadsoft/Metaswitch/Oracle/Cisco your team are, this is a really painful experience. With NFV/SDN, you provision your services virtually and (critically) elastically, and then order services as you need them. Need encryption? Order the service: the SDN controller will automatically route traffic through a newly-provisioned encryption device/VM. And so on.
Net net: SDN is hugely overhyped, but for some orgs, it and its NFV worldview are the future.
Re: makes sense
I'm sorry. While KVM had a surge of activity, Xen is anything but obsolete.
Reading the fine print I see that Terremark's cloud is actually a data center in Virginia. When you have a service that can run your application/service in any one of a number of facilities around the globe, and move workloads and client connections seamlessly between them, then you have a cloud. Until then you just have a large raindrop.