Amazon and Microsoft Need Mainframes
According to Wikipedia, lightning occurs somewhere in the world about 44 times per second on average. A quarter of those lightning flashes strike the ground. Meteorologically, lightning is extremely common. All the more reason to wonder why both Amazon's and Microsoft's customers suffered hours-long outages due to a lightning strike.
I found Amazon's advice to its customers particularly galling: "For those looking for what you can do to recover more quickly, we recommend re-launching your instance in another Availability Zone." Translation: You handle your own disaster recovery (if you can), because obviously we can't.
These are certainly not these companies' first outages. It's rational to assume they won't be the last.
Fortunately Amazon and Microsoft customers have an alternative. They can follow these simple steps:
- Find at least two data centers, physically separated — your own, or someone else's. (Scores of IT service companies, not only IBM, operate mainframe-based clouds. They used to be called "service bureaus.")
- Put a mainframe at each site — your own, or share someone else's.
- Use any of several common, cross-site disaster recovery features available with mainframes, notably IBM's GDPS. Choose whichever flavor meets your particular RTO and RPO requirements.
- Hire competent IT staff, and pay them reasonably.
- Put your applications and information systems on these mainframes, at least for your most critical business services, end-to-end.
- Stop wasting money with Amazon and/or Microsoft.
|by Timothy Sipples||August 8, 2011 in Business Continuity, Cloud Computing |
TrackBack URL for this entry:
Listed below are links to weblogs that reference Amazon and Microsoft Need Mainframes:
Actually, Amazon bought a mainframe back in 1999. Scrapped it in 2000. Of course, this was before the land of IFL's and the savings they offer.
Posted by: Mike Taylor | Aug 8, 2011 1:20:08 AM
This is very sensible advice, especially for startups on a shoestring budget. Can you tell us more about IBM's entry line of mainframes? Can I do this GDPS thingie on a budget of about 25k?
Posted by: Richard Brannigan | Aug 9, 2011 3:47:45 PM
Well said. I find it amazing that Amazon would claim that Availability Zones is the answer. A few months ago, another Amazon outage (in US) took out multiple availability zones. Also, interestingly, at the time, they said that clients need to design for their own DR/BCP. Plus, in both situations, Amazon was blasted for their LACK of communications with their clients. I guess you get what you pay for.....not much.
Posted by: Ken Cameron | Aug 9, 2011 7:53:36 PM
Richard, startups aren't buying Amazon's physical servers or data center building either (which probably costs millions). As is clear in the original post, if you have $25K to spend on GDPS-protected mainframe computing, then you would buy the same business model Amazon has been pitching: cloud services. But, unlike Amazon, you could buy it with an actual, fulfilled Service Level Agreement (SLA) that meets your requirements. Mainframes have been offering cloud services since at least the 1960s, variously called "service bureaus" and "hosting." (Amazon didn't invent anything here; they're just executing badly.) Those mainframe cloud services are available from scores of vendors, as pointed out.
For example, if you're developing applications and want to compile and/or test them on z/OS, IBM offers z/OS cloud access for $300 per month ($3.6K per year). That's not at a GDPS SLA tier, but it's also not $25K.
Ken, I agree. Part of what I find galling about Amazon's statement is their use of the term "Availability Zone" which is now so obviously a marketing term that means nothing in practice.
Posted by: Timothy | Aug 9, 2011 8:46:47 PM
I heard IBM invented hypervisors too, how'd that turn out?
Posted by: Richard Brannigan | Aug 9, 2011 8:59:02 PM
I don't understand your point, Richard.
Posted by: Timothy | Aug 9, 2011 9:11:02 PM
I agree on needing mainframes - have said it often. But the compelling nature of the Cloud is how quickly you can "turn on" a server. Can't do that with Mainframes yet. Fix that and now we're talking. Great blog article btw.
Posted by: Pat Egen | Aug 10, 2011 5:45:18 PM
Pat, a couple comments:
1. Why do you think it isn't quick to turn on a virtual server on a mainframe? It's at least as quick as anything else. (There are no blade provisioning steps, for example.) I don't know where that mythology comes from.
OK, actually I do, but it's still mythological. There are some shops that *choose* not to run their mainframes that way, but that's 100% a human choice, not a technical characteristic. Many shops do run their mainframes that way. And if you want to see every shop move quickly, watch what happens if there's a disaster: if it's a mainframe, hundreds or even thousands of applications and users get re-provisioned very, very quickly. :-)
2. Most z/OSers, in particular, think it's pretty silly to spin up a whole operating system just to serve one new user or application. They certainly could (particularly with z/VM), but they don't have to. They don't suffer from "virtual server sprawl." They run an operating system which supports multi-tenancy within a single instance -- with security, dispatching, and other service attributes fully protected for every user. That's a unique advantage. Said another way, the quickest provisioning of all is the provisioning you *don't* have to do.
Posted by: Timothy | Aug 10, 2011 7:39:13 PM
The comments to this entry are closed.