I’m pondering my poll results from the Gartner data center conference, and trying to understand the discontinuities. I spoke at two sessions at the conference. One was higher level and more strategic, called “Is Amazon or VMware the Future of Your Data Center?” The other was very focused and practical, called “Getting Real with Cloud Infrastructure Sevices”. The second session was in the very last slot, and therefore you had to really want to be there, I suppose. The poll sample size of the second session was about half of the first. My polling questions were similar but not identical, and this is the source of the difficulty in understanding the differences in results.
I normally ask a demographic question at the beginning of my session polls, about how much physical server infrastructure the audience members run in their data centers. This lets me cross-tabulate the poll results by demographic, with the expectation that those who run bigger data centers behave differently than those who run smaller data centers. Demographics for both sessions were essentially identical, with about a third of the audience under 250 physical servers, a third between 250 and 1000, and a third with more than 1000. I do not have the cross-tabbed results back yet, unfortunately, but I suspect they won’t explain my problematic results.
In my first session, 33% of the audience’s organizations had used Amazon EC2, and 33% had used a cloud IaaS provider other than Amazon. (The question explicitly excluded internal clouds.) I mentioned the denial syndrome in a previous blog post, and I was careful to note in my reading of the polling questions that I meant any use, not just sanctioned use — the buckets were very specific. The main difference in Amazon vs. non-Amazon was that more of the use of Amazon was informal (14% vs. 9%) and there was less production application usage (8% vs. 12%).
In my second session, 13% of respondents had used Amazon, and 6% had used a non-Amazon cloud IaaS. I am not sure whether I should attribute this vast difference to the fact that I did not emphasize the “any use”, or simply because this session drew a very different sort of attendee, perhaps one who was farther back on the adoption curve and wanting to learn more basic material, than the first session.
The two audiences also skewed extremely differently when asked what mattered to them in choosing a provider (choose top 3 out of list of options). I phrased the questions differently, though. In the first session, it was about “things that matter”; in the second session, it was “the provider who is best-in-class at this thing”. Where this really became a radically different result was in customer service. It was overwhelmingly the most heavily weighted thing in the first session (“excellent customer service, responsive and proactive in meeting my needs”), but was by far the least important thing in the second session (where I emphasized “best in class customer service” and not “good enough customer service”).
Things like this are why I generally do not like to cite conference keypad polls in my research, preferring instead to rely on formal primary research that’s been demographically weighted and where there are enough questions to tease out what’s going on in the respondent’s head. (I do love polls for being able to tailor my talk, on the fly, to the audience, though.)
For a while now, I’ve been talking to Gartner clients about what concerns keep them off public cloud infrastructure, and the diminishing differences between private and public cloud from service providers. I’ve been testing a thesis with our clients for some time, and I’ve been talking to people here at Gartner’s data center conference about it, as well.
That thesis is this: People will share a data center network, as long as there is reasonable isolation of their traffic, and they are able to get private non-Internet connectivity, and there is a performance SLA. People will share storage, as long as there is reasonable assurance that nobody else can get at their data, which can be handled via encryption of the storage at rest and in flight, perhaps in conjunction with other logical separation mechanisms, and again, there needs to be a performance SLA. But people are worried about hypervisor security, and don’t want to share compute. Therefore, you can meet most requirements for private cloud functionality by offering temporarily dedicated compute resources.
Affinity rules in provisioning can address this very easily. Simply put, the service provider could potentially maintain a general pool of public cloud compute capacity — but set a rule for ‘psuedo-private cloud’ customers that says that if a VM is provisioned on a particular physical server for customer X, then that physical server can only be used to provision more VMs for customer X. (Once those VMs are de-provisioned, the hardware becomes part of the general pool again.) For a typical customer who has a reasonable number of VMs (most non-startups have dozens, usually hundreds, of VMs), the wasted capacity is minimal, especially if live VM migration techniques are used to optimize the utilization of the physical servers — and therefore the additional price uplift for this should be modest.
That gets you public cloud compute scale, while still assuaging customer fears about server security. (Interestingly, Amazon salespeople sometimes tell prospects that you can use Amazon as a private cloud — you just have to use only the largest instances, which eat the resources of the full physical server.)
I’m at Gartner’s Data Center Conference this week, and I’m finding it to be an interesting contrast to our recent Application Architecture, Development, and Integration Summit.
AADI’s primary attendees are enterprise architects and other people who hold leadership roles in applications development. The data center conference’s primary attendees are IT operations directors and others with leadership roles in the data center. Both have significant CIO attendance, especially the data center conference. Attendees at the data center conference, especially, skew heavily towards larger enterprises and those who otherwise have big data centers, so when you see polling results from the conference, keep the bias of the attendees in mind. (Those of you who read my blog regularly: I cite survey data — formal field research, demographically weighted, etc. — differently than conference polling data, as the latter is non-scientific.)
At AADI, the embrace of the public cloud was enthusiastic, and if you asked people what they were doing, they would happily tell you about their experiments with Amazon and whatnot. At this conference, the embrace of the public cloud is far more tentative. In fact, my conversations not-infrequently go like this:
Me: Are you doing any public cloud infrastructure now?
Them: No, we’re just thinking we should do a private cloud ourselves.
Me: Nobody in your company is doing anything on Amazon or a similar vendor?
Them: Oh, yeah, we have a thing there, but that’s not really our direction.
That is not “No, we’re not doing anything on the public cloud”. That’s, “Yes, we’re using the public cloud but we’re in denial about it”.
Lots of unease here about Amazon, which is not particularly surprising. That was true at AADI as well, but people were much more measured there — they had specific concerns, and ways they were addressing, or living with, those concerns. Here the concerns are more strident, particularly around security and SLAs.
Feedback from folks using the various VMware-based public cloud providers seems to be consistently positive — people seem to uniformly be happy with the services themselves and are getting the benefits they hoped to get, and are comfortable. Terremark seems to be the most common vendor for this, by a significant margin. Some Savvis, too. And Verizon customers seem to have talked to Verizon about CaaS, at least. (This reflects my normal inquiry trends, as well.)