Clone it? Sure. Beat it? Maybe. Why not build your own AWS? You can't move without IT companies telling you about the "amazing" new technologies and features they've just launched, how you can't live without them, and what a shock it is that you've managed all these years before they were developed. And of course the bigger the company, the more new stuff they tend to pump out and the …
This article belies an almost total ignorance of the breadth of services that AWS provides. Has the author actually used it in anger or is this article based entirely on reading a couple of announcements?
"Is it true, then, that anything AWS can do, I can do better? No. Yes, there may be a handful of things that are trickier, more expensive or a bit more involved than keeping things internal, but the benefits will generally be worth the effort and cost."
That paragraph doesn't even make sense.
"The last four on prem places I worked had multi-year up-times with no failures."
You must have been working with some unusually competent shops. I used to sell on prem gear. It was rare to have a storage system outright fail, but not terribly uncommon to have some unexplained performance degradation or integration issue (e.g. upgraded firmware on the Brocade switches, now servers can't find storage). When you think about all the time and money companies spend designing, testing, talking about HA/DR, having a guarantee is pretty handy.
"This article belies an almost total ignorance of the breadth of services that AWS provides."
Indeed. Take RDS - yes, you could do it with your favourite DB, bodge some scripting around replicas and backups, bolt on Tungsten etc... but you'd still be trying to make it reliable six months later.
Or Kinesis - anyone who's actually used Kafka in a production environment knows how unpleasant that is.
Or EMR - ever tried installing Hadoop from rpm? Think you can do it in 5 mins?
The list goes on. AWS (and GCS and Heroku) are just so much easier to use than doing it yourself.
Look at how all the big FTSE 100 IT departments and GDS are trying to replicate AWS and dismally failing.
"Look at how all the big FTSE 100 IT departments and GDS are trying to replicate AWS and dismally failing."
Agree. I have seen a number of botched "I'm going to build an X like Google/Amazon" projects. I try to tell them that the could just log on to a console and all this stuff is already built. We'll have it up and running today and it will be way less costly than what they are up to.
"Baffling"? I don't think so. I know what that sentence says.
Overall the article was pro-cloud and pro-AWS, so I don't know why you are making out the author has got a downer on AWS. I thought the message was, there is little you can't do yourself, but it's often a lot easier to let them do it, but you still have some responsibilities. As a regular cloud user, that sounds right to me.
I suppose everyone brings their own baggage with them when reading these articles... :)
Oh right, I think I misread.
"Yes, there may be a handful of things that are trickier, more expensive or a bit more involved than keeping things internal, but the benefits will generally be worth the effort and cost"
From the perspective of this application developer I can't think of _anything_ that's easier to do _properly_ in-house than in the cloud.
Even a simple small office NAS is a PITA to manage compared to Box/Dropbox/whatever.
There are pros and cons to both solutions and it really depends on a clients existing infrastructure, financial position vs their medium/long term business & IT objectives. The article, while brief, tries to cover these challenges and whilst it cannot run into depth all the services that AWS offer, I felt that it was a fairly balanced piece.
What it doesn't cover, is the bit in between. Which has the potential to offer you the best of both worlds. The benefits of on-site infrastructure, with the value of cloud consumption models and services. These will be coming thick and fast over the next 12/24 months with companies like HyperGrid (with Nutanix soon to follow I'm sure) leading the way.
Renting services from AWS or other providers makes sense if you carry out projects that might fail (such as the CEO's pet buzzword du jour concept) or have a fixed lifespan (such as migrating to new services) -- a year down the line you can simply trash the data and tell the provider the monthly contact is at an end. The alternative is to be paying for hardware and premises that you built out for a project but have little use for now (until the CEO hears a new buzzword...)
Did you know you can make your own candles? Get some wax, wick, melt the wax, dip the candles. You can probably make your own wax and wick as well.
Same applies to the pencil.
Why not keep sheep and knit your own clothes?
This article completely ignores the economies of scale. And specialisation.
If Amazon/Google/MS are hosting hundreds of thousands of servers then it makes sense that I can spin up a shared VM on those servers, which will probably not be 100% utilised (the VM, not the server). Heck, I can even shut down my perf test environment when I'm not doing perf testing, and stop paying completely. On prod, I can spin up extra app VMs when I need them using auto-scaling, and then spin them down when demand drops. And then I stop paying. And lots of other reasons too.
The article makes the point that in a business that prefers capital spending, buying your own kit might be a better fit than leasing AWS kit.
But this ignores the fact it isn't just *kit* you have to think about. It's the people that operate the kit. It's the place the kit lives and the power and cooling it needs. It's the software running on the kit. It's the orchestration of services on the kit. It's the software that does the orchestration, and the management of that... and so on and so on.
No wonder we have companies who get into so much trouble, because they think that IT is just about hiring boxes off the cloud or buying them instead.
As an AWS user in a small team, the idea of returning to the days where I had to manage *all* of the concerns that I get from AWS - not for free of course - but for a reasonable fee.... It fills me with horror.
It's a reasonable fee now but,
More people use cloud then less people buy hardware which will probably increase cost pushing more people to the cloud. Large companies will just start making their own hardware for their cloud or purchase hardware companies.
Less hardware being used means less people able to use the hardware as less jobs.
Then the cloud is your only way of working.
I doubt that fee will still be reasonable then.
No.
Ive had quotes recently that make renting a flat and kitting it out look dirt cheap.
Even if you want multisite redundancy.
Leased lines are plummeting in price and the cost of hardware is too. Ive just renewed and upgraded a contract and shaved off 75% of the cost. Plus im now getting commission. :P
Cloud is miles cheaper though. Id much prefer to have my own infrastructure to manage but unfortunately datacentre pricing doesnt allow for this.
In my honest opinion the datacentres are killing the datacentres. This is why they're gradually closing down.
Ive just migrated a client from their racks in Milton Keynes to Azure (yeah, I know, but its a .NET shop) because EasyNet have decided to shut up shop up there.
Thanks to this, ive managed to save my client a staggering £50k a year.
That massive saving is the only reason I've managed to avoid a cut in fees. If the price was like for like im sure id have had to take a hit due to the perceived reduction in IT management overhead. To the beancounters out there, just because its easier doesn't mean theres less work only less obstacles.
So yeah, it's a weird world right now. The DCs pricing is killing the desire for company owned hardware but is sufficiently high enough so as not to kill off engineers.
A junior dev is far less than £50k. Especially .NET ones.
The going rate in London is around £35k salaried. £250 a day contracted short term.
Competition for Jnr Dev roles is high right now in London. Its a buyers market.
I personally have 15 years experience and I am generally only seeing around £400-£450 a day contract rates.
Every business is doing everything to make a profit, but it is still a better deal.
An analogous situation is asking if you can build a car with ordered parts as well as Toyota, Nissan, etc. Could you build a working car? Sure, assuming you have automotive skill. You are never going to get the volume pricing as low as the major auto makers though, even with the automakers profits built in, and they have the process of building the car nearly fully automated for cost and quality. Even if you build your car, now it is on you to do all of the break-fix. You can call the parts suppliers if something breaks, but they are just going to say the problem isn't with their part (see EMC, Cisco, HP, etc support).
or another one is that everyone is already a cloud network user through your teleco provider. For some reason everyone would scoff at the idea laying fiber between their sites instead of using an MPLS or the like from their teleco because the cost/complexity of laying fiber is high, yet they think it is perfectly reasonable to build data centers at tens of millions in cost. Everyone gets the cloud value proposition with respect to their network, just because they are accustomed to thinking of their network as being run by a teleco. The way it has always worked.
All that stuff that apparently is a doddle to stitch together - keep stitching , because its a never ending job of stumbling over bugs, applying patches and updates, creating enhancements, and working out what root cause in a morass of interconnected custom stuff broke the system.
I know a LARGE company that thinks they can build their own cloud infrastructure because (apparently) the others arent doing it right. So far, 18 months in and circa 50 people (this aint no tinpot org) they've achieved zilch, zero, nada in terms of actual computing infrastructure their users can use. Who are instead just buying stuff from AWS and Azure that, well, it just works and is here now, not in 'alpha stage one in 3 montsh time' (its always in 3 months time). Note, this is not on premise implementation they are doing, they are setting out to build a full generic cloud with all the charging, the scaleability, the fault tolerance and so on.
They dont and never will have, the economies of scale and the deep technical expertise, nor the clout that a large cloud vendor has to batter errant suppliers into submission. They'll see this eventually once they realise that several tens of millions of pounds on people to implement "free" software, is more expensive than just buying it off the shelf. Especially since they will never likely ever release anything usable and they'll have spaffed it all away.
AFAIK none of the people in charge of this grandiose vanity project build their own cars or houses, they just buy a ready built one, yet if its computing, apparently its different.
Its not.
This project sounds like it was never intended to generate any business. Sadly, with R&D credits there are investors that pile money into a dead end project just for the tax break.
I feel sorry for the suckers employed by this outfit, they think they're safe.
Lots of good points here, but I would add that the last few places I've worked, the datacentres and comms have come bundled with a list of risks as long as your arm. X isn't resilient, but we aren't pay £y millions to fix. Z hardware is end of life, but no-one can be bothered to pay to upgrade it. Datacentre A is running low on space and power. Datacentre B now has a business that stores flammable materials next to it. With the cloud, all this is not our problem. Add that to the beancounters ability to map actual use/benefit to cost via the PAYG model, and there's a lot that's attractive about the cloud.
Possibly true for AWS. Not sure, don't use them.
Some of the things cloud providers are doing though you just cannot build on premise. Look at Google Cloud Platform. Google uses a 2 petabit/s (yes, petabit) interconnect network in their data centers. You can't buy that on your own from Cisco or Juniper. It was invented by Google. Google owns the world's largest IP network with 200 or so PoPs and caching data centers in every corner of the world (why YouTube videos don't need to buffer). You can't do that on your own unless you want to start laying transoceanic cables. Google uses custom hardware tip to toe for security and cost reasons. You can't do that... or at least not at a reasonable cost.
Also, one of the biggest advantages of cloud is that you pay for actual utilization in near real time, not your once every year peaks. That can be huge. If you are a retailer building out infrastructure because on Black Friday your workload is 10x the average, you have a ton of infrastructure idle most of the year. You never really know where your peak will be either so it is just common practice in infrastructure sizing to take what you think you'll need and add 20%. You don't need to do that in cloud.
Yeap .. Google and Amazon, maybe even MS, have deep deep deep pockets of cash to build just about anything they want using the best new products available .. compared to lagging behind followers players like IBM, HP and Oracle who likely stand up a DC with spare parts waiting for someone to purchase some cloud time.
Most of prvate cloud products really haven't taken off yet.. why buy HW and hire a staff to run it when the entire stack can be outsourced to AWS and pay pennies on the dollar ? Rackspace isn't doing that well managing private OS deployments either.
VMware and vcloud is the largest deployment of inhouse cloud computing. I suspect products like Mesosphere DC will have more success that AWS clones have.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
