Summary:

The cloud is more than just a company buzzword — deciding when and how to implement it is a difficult task for many organizations.

Russell Warman Trader Media Group Bruno Silva University College London GigaOM Structure:Europe 2013
photo: Anna Gordon/GigaOM


Session Name: How Was It For You?

Chris Albrecht
Barb Darrow
Bruno Silva
Russell Warman

Chris Albrecht 00:02

Thank you Om. Thank you Jason. The next one, cheekily titled How Was It For You? It’s going to be moderated by Barb Darrow. She’s the Senior Writer of GigaOM. She’s going to be talking with Bruno Silva, Research Computing Platforms Team Leader at University College London, and Russell Warman, head of infrastructure at the Trader Media Group. Please welcome our next panel to the stage.

Barb Darrow 00:31

We’re back. We have two actual users, or users here, cloud evaluators, which is really refreshing because this is not a vendor pitch. This is lessons learned, priorities when you’re weighing a cloud migration. Bruno, can you tell us what you do and who your audience is and then Russell, and just give us a little snap shot.

Bruno Silva 00:53

Right. As my job title indicates, I’m Research Computing Platforms Team Leader. What we do at University College London is, we basically provide the services that are offered to researchers for high-performance computing, among other things. We tend to call it research computing because it’s actually encompassing. It’s not just the high-performance side of things, but computing in general. We offer those services. We offer training. We design some of the platforms that are offered. We do a bit of R&D work to look ahead and see what people will be requiring in the future. We try to give the best service wrap around these things as possible. We also interface with regional infrastructures, like Center For Innovation, which is conglomerates, South Hampton, UCL, Oxford, Cambridge, and Imperial College. We basically also offer the service wrap for those external platforms. We have some experience there with dealing with vendors that’s gone that way of computing resources.

Barb Darrow 02:05

Russell?

Russell Warman 02:07

I work for Trader Media, and we are responsible for all the Trader dot Code UK which is the UK’s largest motoring web site. We’ve got 11 and a half million unique users that consume as they search for cars, but the main revenue comes from the services that we provide to dealers, so looking at creating products and services that they can use to give them insight into how to price a car, what cars to buy, and then making sure that they get the most out of their connections with the consumer as well.

Barb Darrow 02:39

Both of you guys are evaluating and assessing cloud solutions for various things. Can you talk about what goes into that decision? Maybe what the biggest fear is, making this move, what maybe the CEO’s might be saying, concerns they might have.

Russell Warman 03:03

From our perspective, a lot of it is around reducing cost, is what the CEO is looking for. They want to see a lower operating cost around that. I think in terms of some decision-making around cloud is trying to work out what the differences are between them, and we got to the point where we put a stake in the ground and said, Let’s try one technology and see how it goes. It didn’t work for us, and we ended up choosing another technology instead.

Barb Darrow 03:30

Can you share which technologies these were?

Russell Warman 03:33

Yes. We started off looking at OpenStack, and that was probably about 18 months ago. We got to a point that, because we’ve got some work load in Amazon at the minute, and we wanted to bring it back in house, and the tool that we used for managing that, there was no API that was available. It wasn’t going to be available in the time scales that we were looking. We moved across to CloudStack. We hit a similar problems in terms of trying to get that implemented as well.

Barb Darrow 04:00

Now are you still evaluating that decision?

Russell Warman 04:03

No. We’re going to try again with CloudStack. We were waiting for a later release, which came our earlier this year. We’re scheduled to try that again in about three weeks’ time.

Bruno Silva 04:15

From our perspective, from University College London’s perspective, there’s an aspect of economics as well, so we’re trying to look at other options other than providing everything ourselves and also there’s the technology innovation aspect of things. Looking at the– if you want to talk about definition of cloud and things like this – I know there’s someone who’s going to talk about it tomorrow so I won’t go too deeply into that – but if you look at what cloud really is, we’ve been offering that, institutions and higher education organizations, have been offering that to researchers for a very long time now.

Barb Darrow 04:52

The ability to spin up and down.

Bruno Silva 04:54

Basically the ability not so much to spin up and down but to tap into existing resources that are not handled by the researchers directly. We do so in a way that presents itself as very much an interface with a Unix environment or a Linux environment, whereas the way things are done with public cloud providers is through these API’s, these… it’s this other way of interacting with things that we’re very interested in. It’s not so much trying to outsource, say, the provision of computing cycles, but also looking at an alternative way of offering these computing resources to people other than putting them through the ordeal of going through a scheduler, where they have to wait for their jobs to run, and at the same time, ask them to optimize their codes to shave off, I don’t know, 20% off the run time, when in fact, the time that they’re spending waiting is comparable to the run time. You see what I mean? There’s this user experience side of things that we’re very interested in exploring.

Barb Darrow 05:56

It’s about cost and efficiency, and it’s also about agility and being nimble.

Bruno Silva 06:00

Exactly.

Barb Darrow 06:02

I’m curious, I was talking to one of the cloud vendors last week. She was saying how a couple of years ago it was all about cloud wash. People would attach cloud to whatever product they were trying to sell. Now, it’s all about hybrid cloud washing. I wonder if you’re seeing that and what you think that means.

Russell Warman 06:20

Cloud washing. Well, cloud is a buzz word, I’m afraid. It’s a catch-all term for all sorts of things. I think in the beginning, we had a bit of that. I certainly felt some pressure from senior management that we should be looking at the cloud because people will be asking questions about it. In the end, we did look at the cloud but with a different approach. We just tried to see what it is from what it is, compare it to what we’re offering and in the end we concluded that there’s potential there. We’re certainly keen to engage. We started doing so already with the likes of Amazon, for example. We’re still very much in an evaluation phase, and understanding whether this makes sense or not. From what we’ve seen so far, it all looks very interesting. Hybrid cloud, it makes sense for a number of reasons. It makes sense because it’s a way to have a smoother transition into getting this outsourcing effect going. But it also provides some flexibility because costs vary over time, and local costs, external costs vary over time. It gives you that flexibility. That’s at least our view.

Bruno Silva 07:36

The language has changed slightly, so for the traditional hardware vendors that have now got maybe a public cloud offering, it’s a way that if you start selling hybrid cloud, it’s a way to get traditional enterprise into a cloud. Rather than an enterprise sitting there thinking, How do I get into the cloud, if you start booting out your cloud in your data center and then linking it up to their public cloud offering, I think that’s possibly where–

Barb Darrow 08:04

So you burst up from their equipment in your data center to their cloud somewhere else?

Bruno Silva 08:07

Yes. I think it’s trying to ease that transition through.

Barb Darrow 08:12

Whenever you talk cloud or whenever I talk cloud, I immediately think of Amazon. Its name has already come up several times here. Do you see any other public cloud vendor on the horizon that you would be interested in deploying to, or is it Amazon all the way?

Russell Warman 08:31

We’d certainly be looking to try other public clouds. HP and Google are probably the most mature in terms of what we’ve seen. I think Microsoft have made quite a lot of investment around their platform. Certainly in terms of building out more data centers rather than just being in one place in Europe. They’ve got a few more now. I think in terms of that, and data sovereignty, they’re looking to try and address that, which is quite important for a lot of European organizations.

Barb Darrow 09:03

The whole prism situation, I’m curious how much that weighs on you guys. Do you think about it at all?

Russell Warman 09:10

It’s not really too much of an issue for us, because at the moment our data is here. There’s been some conversation earlier today about people being able to subpoena that data and so on. I think that might become more of an issue and we may get more pressure to make sure that we know where that resides and that we can control it.

Bruno Silva 09:35

UCL being a big, as some people say jokingly, it’s a big hospital with a small university attached. Being a big hospital as it is, there’s the issue of patient identifiable data. I would imagine that there are some concerns around that, protection of data for those people. In general, I don’t think most researchers are very worried that their data is being observed and that their results are being looked at. After all, we’ve outsourced e-mail to Microsoft, and on top of that, they’re forwarding that e-mail to Gmail. They don’t really care much about this. I don’t think that that’s a big problem at the moment. Of course there are political issues around this. It’s questionable, what’s going on. We could talk at length about that. In the immediate future we don’t see a big problem around that. I don’t’ think that that stops us from going to the cloud.

Barb Darrow 10:38

The typical thing I hear about Amazon is that every startup in the universe starts up on Amazon and then I often hear from them, once we hit a certain point and we’re production, Amazon is no longer the cheapest option and end up bringing stuff back in house. Amazon doesn’t like to hear that, but I’m curious if that’s your perception of what will happen with you guys, or what’s happening out there.

Russell Warman 10:59

Definitely. I think we’ve had instances running for part of our business in Amazon for about three years. If you look at the cost of us hosting those instances within our data centers, it’s massively cheaper. That’s part of the reason why we’re looking to build this private cloud with an API into Amazon to bring them home effectively and reduce our operating costs.

Bruno Silva 11:26

It’s the same thing we observe. If you treat the data center as a sunk cost and you just bring in kit and you take power and staff, everything that goes with it, we’ve worked out to about four times cheaper to do it ourselves internally–

Barb Darrow 11:43

Provided you already have the data.

Bruno Silva 11:44

Provided that yes, you have to have that sunk cost. That’s our situation. We have the data center space, so it turns out to be cheaper. The way we’re looking at it is have that as an optional thing to offload some of the queues and do bursting into the cloud and then pull it back when.

Barb Darrow 12:02

You own the load, you rent the spike. Fanous expression from [inaudible].

Bruno Silva 12:05

Exactly.

Russell Warman 12:05

Or if you want to try something new out, and you don’t want the risk of investing a lot of Capex into hardware, it’s perfect for something like that. But I think once you’ve sized it and you know whether it’s going to work for you, and bringing it back in house you can operationalize it, you can manage it more effective within the data center.

Bruno Silva 12:25

Another thing that I would like to see at least from taking an educational sort of view on this, you see software companies offering academic discounts and educational discounts. I haven’t seen yet any initiative to offer something similar for people effectively using–

Barb Darrow 12:41

Amazon [inaudible]

Bruno Silva 12:42

–infrastructure software, right? It is software that they’re offering, so why not get the discount as well to get people in there.

Barb Darrow 12:49

Amazon, student and teacher edition.

Bruno Silva 12:51

Why not?

Barb Darrow 12:52

They say they offer lots of options. One thing, when I talk to customers, I hear issues about vendor lock in. I hear it now in their data center. I hear it starting to come out now about cloud providers. People don’t want to lock in to Amazon. They don’t want to lock into anybody. Is that a real concern?

Russell Warman 13:11

Yes. Definitely. I think we’ve got aspirations to remain cloud-agnostic. If we do use any cloud provider for burst activity, we would want to just use it as commodity and whoever’s cheaper, we’ll just move the workload where it needs to be moved to. That’s where we want to get to.

Bruno Silva 13:36

If we move to a scenario definitely where we’re outsourcing most of the compute, we certainly would like to be able to shift our workloads according to cost. It’s the same thing.

Barb Darrow 13:47

Basically point them to a different cloud.

Bruno Silva 13:48

This issue of API’s and things like this, standards need to be defined and it’s difficult to have a standard that’s a associated with a vendor because it’s one of the methods they use to keep–

Barb Darrow 14:02

To lock you in.

Bruno Silva 14:04

It’s not just in this scenario. For example, GPU computing is a big thing right now. Nvidia came up with a language called CUDA, to program for GPU’s, to do computing on GPU’s. Well, that’s a proprietary language, so you have loads of researchers now that are very familiar with programming in CUDA, and they’re locked into that because it’s difficult to get into the language, and once you’re in there, that’s what you know, and that’s what you want to do. You end up just buying Nvidia hardware. It’s something we would like to avoid, but…

Barb Darrow 14:37

Sometimes you can’t.

Bruno Silva 14:38

Sometimes you can’t.

Russell Warman 14:40

Sometimes it’s inevitable.

Barb Darrow 14:42

If you were to look out there, I’m curious about your DevOp Shops, or are you DevOp Shops? Are you moving in that direction?

Russell Warman 14:50

Our aspiration is to be there.

Bruno Silva 14:51

Yes, we’re moving in that direction, I think, but strongly.

Barb Darrow 14:54

And the skill set’s there for you guys? Are you training your own people?

Russell Warman 15:01

We’re having to train our own people because we find recruitment based on where we are in the UK pretty hard to get people with those skills, because we’re based up in the Northwest. There’s not the same amount of talent in that area, is there as perhaps in London.

Bruno Silva 15:19

We have a slightly different concern, which is we can’t hire more people at the moment.

Barb Darrow 15:25

That’s a common thing.

Bruno Silva 15:27

It is, so we’re very focused on training our core staff and we’re trying to get everyone to first understand the concepts around it? What are the implications? Trying to automate as much as possible and get rid of any manual work script things, program things, devolve.

Barb Darrow 15:45

Work fast, fail early, retry.

Bruno Silva 15:47

Exactly. Devolve some things to developers to the researchers that are trying to develop code, rather than doing it all ourselves. It’s been working. I think we’re developing that culture and it’s starting to show.

Barb Darrow 16:01

We have a minute left. Would anyone like to ask a question? I can’t see anything. Go ahead.

S5 16:11

I know that some Amazon services because they’ve got such a history. They know how to protect their data and they know how to maintain, up and running. Doesn’t risks on you, switching back from Amazon to have to maintain and protect it properly against attacks [inaudible]?

Russell Warman 16:31

We run the majority of our commercial web sites ourselves now, anyway. A lot of those things that we’ve got in place, like DDoS mitigation,WAFT, it’s already on our infrastructure, so it’s just a case of scaling it up to support what bit of the workload that we’re bringing back. We’ve already got teams of people that are looking at those things specifically. It’s not a new challenge. We’re already trying to address it.

Bruno Silva 16:58

In our case, UCL is an organization that is making– well first of all, because it’s being driven by funding bodies in the UK, but it’s making an explicit bet on protecting UCL data as an asset, so there’s a team of people that are specializing in making sure that we’re able to keep that data at UCL. I’m sure that the archiving aspects of things might end up being outsourced and end up in the cloud, but that’s not so much active data. Active data is staying very much at the core of UCL. Actually we’re working with the DN on this and it is something that it’s strategically important and that’s one of the blockers to full-on move into the cloud, is this need to have this data locally. As you know, pushing data back and forth into a cloud–

Barb Darrow 17:44

It can be expensive.

Bruno Silva 17:45

–can be very expensive.

Barb Darrow 17:47

That’s it. That’s all. These guys, you might be around for later, if someone has questions afterwards? Thank you very much.

Chris Albrecht 17:58

Thank you Barb, and we’re moving into our afternoon break. We hope you’ve enjoyed the session so far. We’ve got a one hour break. Please visit our sponsors. There’s also some workshops going on. The Nuage Network Sponsor Workshop, The HGST Sponsor Workshop, and the DataDirect Sponsor Workshop. Nuage is happening in Sidney suite on level one. HGST on the Beaumont suite level one, and DataDirect Networks on the Harpley suite level one. Stop by the GigaOM Research Table, find out more about GigaOM research. Breaks and refreshments next at this level, and we’ll resume the general session at 4:30. 16:30. Thanks a lot everybody.

S6 18:47

Introducing Green Cloud. With Compute Cloud and Storage Cloud. Highly scalable, easy to use, cost effective renewable energy powered cloud services. You can host your web site on a renewable energy driven cloud. Use Compute Cloud to spin up a virtual machine in minutes. Store data and increase scalability through storage cloud, manage everything with ease through our interoperable dashboard. All while minimizing your carbon footprint. Log on for your free trial today.

You’re subscribed! If you like, you can update your settings

firstpage of 2

Comments have been disabled for this post