- Want to liven up a boring conversation? Bring up OpenStack appliances.
That will get people talking. In an enthusiastic manner. With vigorously-defended opinions. The appliances talk was by far the most interesting one I went to all day. And you’d think that with the panel consisting entirely of people whose companies offer or are planning to offer OpenStack in an appliance or something very similar and pre-packaged, it would be a one-way conversation. But not so. The audience asked these guys hard questions.
What was offending the anti-appliance contingent?
Mostly vendor lock-in. Early in the conversation the panelists had spoken about how there really is no lock-in with OpenStack. But it was quickly pointed out to them that if you purchase an appliance, you’re pretty much stuck with the vendor that sold you the appliance, and therefore quite clearly locked-in. (No pulling one over on this audience.) Leave them and you’ve got no support for this physical box. It’s not like simply switching distros.
Panel reply: They didn’t really have one. One panelist debated what “lock-in” really means, and claimed that lock-in can be good if it means you’ve found the best solution out there. Another conceded, “it’s not for everyone” (appliances). And a third pointed out that they built their solution with a product you can do other things with, so that if you don’t want to use the appliance anymore you can redeploy the server. But it really seemed like no one wanted to state the obvious: Of course you’re locked-in.
Also the fact that appliances become obsolete quickly. “What about the fact that the software is a moving target? How do you prevent the appliance from becoming obsolete?” wondered one audience member (to the nods of many others).
The panelists had a better reply for this one: Automated testing. They do that for the upgrade every six months. Then recertify the appliance. Of course, it was immediately pointed out that the whole upgrade cycle they just outlined is at odds with earlier comments about lock-in. But everyone just smiled and moved along.
So what did everyone agree on?
That appliances are a great solution for getting started quickly. They make OpenStack more consumable because they take the work out of standing up the cloud. It’s racked and stacked quickly, so you get the benefits more quickly—even if you don’t employ an army of OpenStack engineers. As one panelist put it “Appliances are a bridge to OpenStack for the mere mortal. The economics favor repatriating your workloads to a private cloud, and this allows them to do that.”
Well put, Sir.
They also agreed that whenever you package something up, you limit your flexibility. That you might make Day 1 with your cloud much easier, but you aren’t really taking care of Day 2 onward. Which brought us right back to the lock-in argument above.
As I said, if you want to liven up a conversation, at least in a roomful of Stackers…
- Cost-savings are still a major driver for choosing OpenStack.
We’ve all been talking for a while about how cost savings are the least important reason for moving to the cloud. And that may be true for the developers who want that self-service, instant-access-to-resources experience (as well as for forward-thinking executives that can see how that will benefit their companies). But when it comes to OpenStack specifically, I’m still hearing a lot of customer stories that illustrate insane savings. They start with statements like “I was told I couldn’t spend ANY more money than I had the year before, but I had to find a way to store three petabytes more data—and I did it with OpenStack.” Or, “Success for me was making it (his company’s new private cloud) better than Amazon (which this customer had been using). With OpenStack we did it for 1/3 of the price and gained 2x the performance.” Or “Our company will save $1M/month on hard drives because of an OpenStack feature that was recently developed.”
It may not be the lead story anymore, but it’s still an interesting angle.
- Walmart has built a pretty big OpenStack deployment, and they have some data to pass along.
- There are only 15 people on their OpenStack team (in case you imagined an army).
- They don’t use block or object storage.
- Like others at this conference, the speaker said changing the company culture has been the biggest challenge: “Some people have been there a long time and done things a certain way. An enterprise way. They don’t get clouds from the get go, so you have to change that. Have to change hearts and minds. If you don’t change how they think about what you’re doing to their DC or apps, it will be an uphill battle the whole way. An executive with a big hammer on your side helps, but it doesn’t solve the problem. People have to want to do it. Not just do it because they’re forced to.”
- They iterate. They deployed a cloud. It wasn’t perfect. So they pulled back, leapfrogged over. Changed what they did with hardware. Now changing the way the network works.
- Instant access to capacity has changed the way resources are consumed at Walmart. Meaning, they get consumed even faster than they ever were before.
- They still have some non-cloud apps that require “pet care,” and cause constraint problems in the cloud. “We treat them like pets we don’t like very much.”
- They contain failure by deploying multiple clouds.
- They use a PaaS layer to glue everything together (OneOps .com).
Oh great! I found your blog when I was looking for a different sort of information but
I was very happy and glad to read through your blog. Thanks for sharing.
Thanks Ali, any further comment on exactly what Walmart had to change culturally? I get the obvious more to consumption based models and devops as typical call outs – but keen to understand some details if available?