AWS v Oracle: Mark Hurd schooled on how to run a public cloud that people actually use Amazon's AWS infrastructure boss has slapped down Oracle co-CEO Mark Hurd after the latter boasted that Big Red needs fewer data centers because its systems are, apparently, twice as good. Writing on his personal blog this week, James Hamilton, an Amazon distinguished engineer, said the suggestion that Oracle can compete with …
... is always, always, always an amateurish, blinkered, cheapskate, short-sighted, fundamentally stupid thing to do. It is, as such, the almost exclusive province of the inexperienced, bean-counters and senior management (and of course, politicians, who somehow manage to embody all the vices and weaknesses of humanity in one shoddy package).
And yet, when you are as old as I am, a kind of wry despair sets in as you watch each new generation sprout its own collection of boneheads who, when not eagerly snuffling after the latest vacuous fad (be it cloud or CRM or AI) and proudly strutting their "knowledge" of "innovative new thing" via endless Death by Powerpoint, confidently assert their brilliance by cutting costs—two of the most fashionable tactics being, One, to discard experienced, knowledgeable staff, and Two, to cut away the "dead wood" of backup, redundancy, testing, failover, recovery, distributedness, re-resting and all the things one might classify as "engineering common sense".
The amateurs should never be allowed to make decisions, but as I'm sure El Reggers are well placed to know, IT is infested with cowboys. The UK is a particular problem—Germans tend not to tolerate this kind of nonsense—and the US isn't far behind. The proportion of arrogant bullshitters spouting terms they read in the trade press yesterday is positively frightening.
Bean-counters ... well, ugh. When you love beans, I guess they are the only thing you care about, and it's a short step to seeing the world only through the rails of an abacus. Perhaps they are the one group we should have some sympathy with; they can't help the blinkers, and probably feel more comfortable with them on: saves having to deal with messy, smelly human beings.
As for senior management—here is where most political vices accrete like a kind of evil booger, and the higher up you go, the ranker the canker. A combination of dishonesty, greed, varying levels of sociopathy and the political DNA that values style over substance, propaganda over fact, is the most fertile possible ground for mendacious short-sightedness and eventual disaster.
The same mental and moral weakness is visible again and again. All part of the spectrum of folly. Whether it's DC-10 cargo doors; Shuttles killing entire crews; idiots advocating adding yet more runways to already critical airport nodes like Heathrow; or a mild enough example from this bonehead at Oracle ... people just never learn. At least Oracle's transparent moonshine won't kill people; let's be grateful for that.
"At least Oracle's transparent moonshine won't kill people; let's be grateful for that."
I'd have to respectively disagree with that point. Within the Health Services (talking NHS in particular here) there are a number of Oracle databases in use that have potentially life threatening consequences if not accessible.
Not sure how that may tie into their 'cloud' strategy but I'd not like to have to put it to the test to find out!
"there are a number of Oracle databases in use that have potentially life threatening consequences if not accessible."
That sounds like a particularly foolish bit of design.
Particularly as the license explicitly says that the Oracle database not designed for situations "that may create a risk of personal injury".
Particularly as the license explicitly says that the Oracle database not designed for situations "that may create a risk of personal injury".
Oracle have a whole page dedicated to their healthcare products.
I find this comment hilarious! "I also would argue that “speeding up the database” isn’t something Oracle is uniquely positioned to offer"
Seriously? since when does AWS have any Database experience and especially making it faster? Oracle has been in the Database business for 40-years now and developing its own hardware for last 6 years and has published 50+ world record benchmarks running Oracle DB. Majority of the world records for Database are from Oracle. Oracle has over 50% marketshare of the Database market when you combine Oracle DB, MySQL and the other Databases in the Oracle portfolio. AWS has what %? Are they even on the radar by Gartner/IDC (for Database)?
AWS platform is 100% virtualized on bare-vanilla x86 infrastructure. As we all know, virtualization has overhead upwards of 30%. You can't even get bare metal from AWS so you're forced to virtualize and even have noisy neighbors.
No one that’s serious about Database performance will run on a virtualized environment. Get serious AWS and publish some benchmarks if you want to prove your statements! This is hog wash otherwise.
> ... since when does AWS have any Database experience and especially making it faster?
You mean the multiple database product's they've been providing for several years now, the thousands of man hours of development, debugging, optimising, and supporting those database products?
> Majority of the world records for Database are from Oracle.
Including such amazing records as "least effective per $ spent", "least effective support" (by several measures), "most loathed database", and "most loathed database company".
But, Oracle seems to be paying your wages at least, so you're probably ignoring those. :)
"Presumably he means overhead on your licensing costs - Do Oracle actually recognise virtualisation in their DB licensing yet?!"
Sort of, I had to look into this for work and basically:
1 Oracle licensing assumes that a core is dedicated to that VM
2 unless you have Oracle's VM software configured in the correct way than HT threads count as Cores so you must turn them off leading to a 30% drop in performance.
I took one look at the above and opted for bear metal servers.
.
" and has published 50+ world record benchmarks running Oracle DB"
Did have. SQL Server has most of the records these day.... Plus about 1% of the security vulnerabilities.
"Oracle has over 50% marketshare of the Database market "
Only by $. MS SQL Server wins hands down for installs + clients.
"As we all know, virtualization has overhead upwards of 30%."
Maybe it does on Oracle's platform. It's about 1-3% on Hyper-V / Azure.
".....No one that’s serious about Database performance will run on a virtualized environment....." And therein lies one of Oracle's biggest problems - they are convinced they are still fighting the Big Iron Database War of a decade ago. The vast majority of cloud customers are not mega corporations, indeed one of the highlights of how cloud has developed is that it really developed from the SMB market. Little web-based software shops in particular, seizing on the cost-savings of using cloud-hosted services so they didn't have to pay for in-house systems and maintenance. Those type of companies were very happy to go with virtualised environments, and the absolute screaming end of database performance was way down the list of their requirements compared to availability, low cost and low latency. AWS cornered the market because they met those requirements for the SMBs whilst having the stretch capability to also provide more performant solutions for the corporations as they moved in after the SMBs. Since Oracle has always had an awful time selling their high-priced software to the SMB market they have always been poorly positioned for the general cloud market that AWS has dominated.
Oracle really needs to clean out all the old management dross like Hurd and Fowler with their blinkered view of the market and get a real cloud CEO.
Maybe it's because fast-growing, real world global scale businesses don't give two sh*ts about artificial benchmarks or pay-to-play shills like Gartner. Call us back when Netflix moves from AWS to Oracle cloud. Oh, yeah, they tried that old skool vertical scaling and declared in 2014 (!) "it doesn't work" https://www.youtube.com/watch?v=uH8T7JMzloM
All the comments here are very correct that modern businesses don't need Oracle at all. Google, Netflix, Apple, Uber, AirBnB have all figured out how to achieve huge scale with globally distributed, redundant and scalable databases. And, by-in-large, they have made those tools cheaply or freely accessible to everyone. And no one who is serious about large scale systems will ever run anything bare metal ;-)
Why the hell would you not run DBs on Virtual machines?
30% overhead? Not seen those kinds of overheads for years.
90% of DBs I see out in the marketplace, do not require anywhere near the amount of
compute capacity available on modern platforms with 18+Cores per Socket in a two Socket server.
If you have the kinds of requirements for 8+ Sockets for a DB then you probably want to run the DB on Prem anyway or at least in a private/sourcing deal as no cloud can provide the kind on SLAs needed with such a DB which is likely to be business critical. (Or the company is crap at archiving data)
"Oracle has over 50% marketshare of the Database market when you combine Oracle DB, MySQL and the other Databases in the Oracle portfolio"
Why combine it. MySQL isn't an Oracle database, it just happens to be owned by Oracle. It can be run without licence by anyone on any hardware or from any cloud provider.
"No one that’s serious about Database performance will run on a virtualized environment."
Orcale cloud is virtualised - no-one serious about databases will use Oracle Cloud - which is Oracles *highly* recommended option?
"Majority of the world records for Database are from Oracle"
Most people aren't running highly optimised world record, money no object boxes and the tests are never 'real-world'. So if to get the fastest database performance you have to spend 3x amount on licensing, 2x amount on hardware, 1.5 x amount on staff to get 2% better performance then maybe that wouldn't be a valid argument for most people. Also a number of organisations nowadays are finding their data suits a non-relational model better. Therefore by organising it this way, where appropriate they could see a 10x performance increase on a fraction of the cost of an Oracle RDBMS.
Oracle is a top contender to this day because of it's legacy Network DBAs, legacy software and third party software deals. It is extremely difficult to transfer to a new DB backend for a very large, highly optimised system. Combine that with the fact that all the staff will be selected, trained and knowledgeable only about Oracle and then chance of a switch out is rare. That is changing as the newer generation of DBAs, tech, and companies are coming through.
If Oracle actually liked its customers, its staff, its sales agents and didn't try to screw them over whenever it was the end of the financial year then they might have a better rep. But I predict that if Oracle are still reliant on just their DB then their time is numbered. They won't die in a flash, they'll lose new customers and their existing customers will run their systems until they no longer need them or no longer have a company.
Oracle people say some idiotic things, such as we don't need as many datacenters. But if you want to run Oracle in the cloud - they are the best choice. We've tried the others - and I am not impressed.
Can you run a large Oracle database as quick on AWS or Azure as on Oracle's cloud? No. Not even close. For small lazy databases however, they are fine.
Want exadata? Of course nobody else has that, they are just generic white box machines.
How about RAC for full redundancy? Not possible due to a lack of hardware config at AWS and Azure.
In Memory? Well, to make use of that you need a large memory computer for that, and their's are smaller.
Of course the whole cloud problem is that to get the best performance, you really need to have everything IN the same datacenter. So if you have a mix of Oracle, SQL Server, HANA, Hadoop, Mongo databases as well as Java and .Net app servers - you have to pick one and suffer for anything that isn't native to the provider. Or... you could just run them in-house and configure them optimally yourself and save money, and not have to worry about them holding your data ransom.
Not entirely on-topic, but Oracle really do need to spash some cash on hardware to support their own services. I've recently started using Oracle Linux and attached a server to their Unbreakable Linux Network (ULN), basically their version of RedHat's RHN/RHSM. In short any interaction with the ULN repos takes an ice age and is completely unusable. As an example, a yum repolist all took almost 30 minutes(!). On a comparable RHEL server talking to RHN/RHSM the same command takes upto 30 seconds.
The reason for this is that Oracle host ULN on their own hardware/network, which is obviously not fit for purpose. More so, when you consider that you have to pay for a licence/subscription/support to access ULN. Ironically the bulk of Oracle Linux is freely available (and at no cost) and is availble at yum.oracle.com which is hosted by Akamai and is fast enough to be usable.
Hmm. Let me see here. AWS is claiming the big "3" Cloud vendors spent $31BN on CAPEX in 2016? FYI. This CAPEX expenditure isn't just on CLOUD !
So is Hamilton claiming Amazons $6.74BN CAPEX spend in 2016 is 100% for AWS and for its customers and not also to support AMAZON? Yeah sure, you belong in marketing... Amazon the retailer uses AWS as its IT backbone to support AMAZON Prime, AMAZON video, etc and up until a few years ago, AWS wasn’t even separated financially.
And Amazon retail growth worldwide needs a lot of IT & warehouses which demand a lot of CAPEX. But of course, looking at the AMZN financials, AMAZON the retailer does not break out CAPEX spend for AWS? Makes it easier for all the armchair analysts to refer to this big false number. My take, ~50% of AMAZON CAPEX going to AWS buildout for *external CLOUD customer use*. Anyone here can prove me wrong?
And how much of Microsoft CAPEX spend of $8.9B in 2016 went to Azure/Cloud vs CAPEX needed for LinkedIn, Skype, Xbox on-line, etc that has nothing to do with Cloud SaaS/PaaS/IaaS? My guess, 60% of that # going to Azure. Again, Microsoft doing such a great job hiding Cloud business in its "Intelligent Cloud" category making it impossible to separate it all out. No one even knows what Microsoft's AZURE revenue even is! Its all a guessing game.
And then Google, yeah they are the gorilla of the infrastructure world, needed for WW Google Search and especially YouTube. Google Capex for 2016 was $10.9BN. So how much of that spend going to enterprise cloud services-GCP? My guess, less than 50%.
Oh and by the way, I count $26.64BN in CAPEX for these big 3, almost $5BN short of the claimed $31BN. Who did the math here? $6.74BN+$8.9B+$10.9BN = $26.64BN
Its what I call Cloudwashing!!
So Oracle's $1.7BN CAPEX spend, although not as high as the Top 3 is currently spending *on Cloud*, are still quite strong investments considering Oracle has only been doing IaaS for a little over 2 years. Takes time to buildout and of course gain momentum. Amazons been building out IaaS for over 10 years since 2004. And looking at their financials history, it was only after 2011, that Amazon CAPEX exceeded $1BN/year and their first 5 years of AWS, invested ~$1.3BN Total in CAPEX, far less than Oracle's 2016 one-year CAPEX spend!
Heres the CAPEX substantiations:
http://www.marketwatch.com/investing/stock/amzn/financials/cash-flow
https://www.microsoft.com/en-us/Investor/earnings/trended/capital-expenditure.aspx
http://talkincloud.com/cloud-companies/google-ramped-data-center-spend-2016
All those big data centres filled with IT stuff must depreciate awful fast...what happens if I add Amazons depreciation and capital expenditure and take a chunk off for Amazons business, multiple by 3 (for AWS/Azure/Google), does the US$30 billion number look right?
And if I go and look at Microsoft and Google, they seem to have an enormous depreciation bills as well...
Far more recent than that!
https://www.theregister.co.uk/2017/03/01/aws_s3_outage/
Amazon might try some alternate reality and claim it's not a 'regional failure' but if you had any system that relied on S3 then you were SOL. AWS/Azure/Google are all very interesting and useful tech but reliable and cheap aren't words that can be used to describe them.
Wouldn't a regional failure require all DC's in the region to fail before the region is considered as failed?
Yes, AWS have had failures of individual DC's but part of Mr Hamiltons argument against Oracle cloud was that relying on a single DC would always (given enough time) result in some event that leads to a failure which is why AWS aims to provide 2+1 DC's per region.
This post has been deleted by its author
"AWS aims to provide 2+1 DC's per region"
Not looking good for London, Frankfurt, Mumbai, Canada and Seoul then. Although AZ != DC it is for any deployable architecture.
Only having 2 AZ makes a properly consistent database unable to recover from a partition caused by a single AD failure (something I discovered by experience, unfortunately)
"... toolchain to build and run Go code on the Xilinx UltraScale Plus FPGAs attached to F1 virtual machines. That's much nicer than fooling around with hardware languages to accelerate bits of your codebase in silicon."
Hey, whatever runs up the energy bill. Go green...pfff, just Go. Now if we can read about putting 1000lb anvils on F1's for increased control (of something you're not qualified to control). IKEA standards applied.
Just pay someone who knows more about it than you. Stop pretending and help create jobs that actually have benefits. Cheap breeds cheap.
It was a horrible expression of choosing more abstraction is not the way to go with resouce crunching... I was on tbe phone while posting.
Basically everytime something more powerful comes along, people don't want to pay people who know how to use it, they settle for abstraction that is too high. In this particular case, I'm not sure I'm correct... I would delete the post, but I meant it at the time.
Mr. Hurd's comments read like the end of the Simpson's Lemon Tree episode, where the residents of the other town lost the lemon tree, but felt good about it--"because it was haunted".
It basically sounds like posturing, to convince people suffering from Oracle lock-in that they should not get too restless about being locked in. Get a real cloud, Mr. Hurd.
Sounds like you should jump out of the frying pan directly into the fire?
This statement says it all from here: http://fortune.com/2015/10/08/aws-lock-in-worry/
"People talk about lock-in with Oracle. But data gravity makes lock-in worse with Amazon," he noted. Data gravity is a tech term that means that once data is inside a given repository, it is difficult and expensive to move it. "With Amazon, there are fees to do that and they're hidden until you try to do it."
So who here believe that AWS, which began life about 2004 or so, has completely overhauled their IT infrastructure with the latest in technologies across networking, storage, compute, SW, etc? Being in the datacenter business, I find it hard to believe that many of the architectural decisions AWS made back in the early 2000's has been completely overhauled and if AWS were to build it out all over again, clearly wouldn't do what they did 13+ years ago? The question today is are they experiencing the same datacenter lifecycle pains we've all experienced the last 10 years and why we are starting to see some more critical services (and regions) like S3 go down?
but to respond to a much smaller competitor in Cloud, as Oracle is, tips your hand that you may actually see real competition.
We'll see how it goes long term, but Oracle has more experience in the Enterprise Market and AWS has more experience in the consumer market -- with some notable exceptions for both. My guess is that they'll both do well in different markets with lots of overlaps.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
