Posts by Paul 14 29 publicly visible posts • joined 8 Nov 2011
I can't see many significant Anti-AI OSS projects surviving the onslaught of security vulnerabilities being identified by... AI.
It was enough of a challenge for OSS maintainers to keep up already, where are they going to find the time? And the vulnerabilities remain real whether you like AI or not!
"the market will forgive AWS this time"
Maybe in the US - but I got really strong vibes from my network in the UK and Europe yesterday that senior people are VERY unhappy that their core business depends on infrastructure on another continent working correctly. Yesterday workloads in other AWS regions - that AWS makes a big song and dance of supposedly being isolated - were impacted by the US-East-1 issue.
Unless AWS makes serious moves to solve the US-East-1 dependency problem, they will lose customers outside the US in droves. Execs and governments care about digital sovereignty. They want to know that they'll still be able to operate if Putin shreds transatlantic cables - AWS demonstrated yesterday that they're not up to that challenge, but worst of all, they eviscerated trust among senior people who believed the region-independence narrative.
This isn't true. Data centres need massive amounts of connectivity - to end users, other data centres, and global internet exchanges. Laying fibre costs per mile, and longer fibre distances also chew more energy. So economically (and environmentally, in terms of energy usage) it's best to cluster data centres together, and closer to population centres.
Not really a mystery, it's all about connectivity and geography, and less to do with latency and more to do with bandwidth, cost and agglomeration.
The London area starts with about 10% of the UK population, all now keen on bandwidth-hungry gaming and video streaming, but previously all keen on the as-was bandwidth-hungry applications of the younger internet. Fibre installation ultimately costs per-mile, so putting the compute closer to the population reduces bandwidth costs. Then this compounds because putting compute close to other compute (some of which it will inevitably need to connect to) also reduces bandwidth costs. Then you add in that this attracts more fibre investment and bandwidth to the same locations, including international connections, and things compound further. Skills availability is another feeder for agglomeration, as is London's position as a popular connection node between America, Europe and Asia.
Functionally the internet appears to be geographically agnostic, but we know that in reality getting bandwidth longer distances to the back of beyond costs substantially more than around major cities. Not only is fibre installation cost per-mile but more energy is required on an ongoing basis to send more data over longer distances. So whilst data centres anywhere there's power available are in theory possible, in practice the bandwidth cost is more of an issue than you might think.
As time moves forward the Moore's law acceleration of compute vs the rather more linear acceleration of connectivity suggests we might see more "edge" compute distributed closer to smaller population centres, but that remains to be seen, and it's hard to see how the agglomeration effect of global hubs like London (+ Singapore, North Virginia, Amsterdam etc) can be avoided.
You want to put your fingers all over a public touch screen and THEN eat McDonalds food with them?
Ewwwww.
I know people already do this in the restaurants but still, ewwwwww.
Use the app on your own phone at least!
These news articles report on two things -
1) McDonalds ending its partnership with IBM to provide this tech
2) A small number of incidents of said tech producing comical results - note that these seem to be the same handful of incidents in every article
There is implication that these two things may be linked, sure, but there is no statement from McDonalds or IBM that they are.
The fact that the trial has been running for 3 years, being expanded from 10 to 100 restraurants for the last 2 of them, but only a small number of examples exist of it going badly wrong, implies to me that it's actually been working quite well for the most part. And even if there are a handful of issues, the normal approach would be to look at those to improve the tech, not abandon it altogether.
So I suspect the reason for ending the trial is more commercial than technical, and these incidents have nothing to do with it.
And the cheery conflation of this announcement with the incidents by the news media is the usual distortion in pursuit of clicks - or as it was once called, sensationalism. Or propaganda.
Was security benefit the primary reason for NAT adoption in the 1990s? No - that was address availability.
Was there some security benefit in the 1990s to widespread NAT adoption? Certainly, particularly for unprotected home users with early Windows versions. NAT would probably disable the original Back Orifice exploit, for example.
Was this security benefit flawed? Of course it was. All security has flaws, but this especially as security wasn't the main objective of NAT.
Has NAT ever provided the same security benefit as a properly configured firewall? Of course not.
Did/Does NAT add any additional security benefit if you have a properly configured firewall? Probably not*
Does NAT provide the same degree of security benefit today as it did in the 1990s? No; since NAT today is commonplace any modern exploit will expect it and work around it.
Does using IPv6 without NAT mean greater security risk than IPv4 with NAT? Not really, because it's only something like the difference between a wide open door and no door at all. A proper security gate - ie a properly configured firewall - is an essential way to mitigate security risk in either case.
Was/is there a danger with people believing NAT provides more security than it really does? Absolutely, with bells on.
*Some might argue that NAT provides additional "security through obscurity", which despite being a much maligned idea, is still put forward by some analysts as having some marginal value - the argument being "why make things easy for an attacker when you can make them complicated". The counter to that is that making anything obscure for an attacker probably also makes it complicated for your own people, which can increase other risks around human error, response times, and limitation of resources - ie time spent working with needlessly complicated configurations could be better spent on more effective security measures.
I've used BlueJeans, Teams, Zoom, WebEx, Chime, Meet and most of their predecessors.
BlueJeans failed for 2 reasons - the biggest was poor scalability at the start of the pandemic. Whereas Zoom had a scalable hybrid-cloud software platform that could overnight meet the demand to go from 10s of thousands of meetings a week to 100s of thousands of meetings a week, BlueJeans just fell over and wasn't available for anyone, so people switched to Zoom and never came back.
The other reason is more about what Zoom did best of all - first-time User Experience. Even though BlueJeans was technically pretty good once you were up and running, getting the software and getting onto a meeting was a faff for ordinary not-technical people. Zoom is where it is because in 2020 you could send a single link to your elderly relatives and get to talk to them.
As ever UX - top quality, sustainable, consistent, scalable UX - is what makes or breaks.
IR35 is only "necessary" because dividends are taxed at a lower rate than earned income.
Or putting it another way, sitting on your a*se owning a piece of paper because you were rich in the first place means paying a lower tax rate on income for doing nothing, than anyone earning money actually doing some work.
IR35 further punishes normal people simply choosing to operate as a small business rather than an employee, bu stillt lets big multinational businesses get away with all manner of intricate tax avoidance schemes.
This system stinks.
I can see the quandary here, which is that these facial recognition systems will continue to be inaccurate without more training data.
But the easiest way to get more training data is of course to deploy the system in public.
So those arguing that inaccuracy is a reason not to deploy the tech in public are missing the point, or perhaps being disingenuous; the tech will never become more accurate without public deployment.
The trouble with focusing the debate on accuracy is that it won't be a problem forever, you're only delaying the inevitable deployment of a system trained with data gathered elsewhere or by other means.
If you are opposed to seeing any kind of live facial recognition on the streets, accurate or not, you need to have the courage to argue against it from a morality/privacy standpoint rather than focus on accuracy, ethnic bias or any other problems that are inherently solvable.
Barclays claim they've ditched all their legacy but still need 20,000 IT staff for their new world cloudiness?? Sounds like BS to me.
These banks are awash with legacy tech in every nook and cranny; the fact that execs might really believe it's all new, shiny and cloudy is terrifying. Especially when they decide to slash their headcount without any grasp of the real legacy stuff these people are maintaining. Sticking legacy apps with waterfall update cycles in EC2 doesn't make them any less legacy! If you're not doing full software defined infrastructure with CI/CD delivered apps, you're doing legacy.
Personally I reckon it's worth splitting your money across multiple banks and putting at least some in Starling, Monzo etc - I think in terms of banking IT failures the worst is yet to come, and it won't come from the new kids who are doing cloud native and know how to do it resiliently.
Some points in response -
1) 5G has been designed as suitable for mm bands, yes, but ultimately like 4G it will be rolled out to all mobile bands. 5G at 800MHz will for example be very nice for rural locations, and could actually remove the need for expensive rural fibre infrastructure.
2) Unlike 4G, 5G incorporates a whole new radio transmit/receive architecture making more of bandwidth at any frequency.
3) 5G is designed to operate in tandem with 4G to increase bandwidth. A likely scenario is a handset outdoors using 4G at lower frequencies to transmit, whilst using both the 4G and 5G at higher frequencies to receive, this providing greater bandwidth to the individual and more capacity overall. This way, moving about, passing trucks etc will simply affect the bandwidth you're getting, not the whole signal.
4) Notwithstanding the above,. the most compelling case for 5G today is indeed to provide capacity in cities where 4G is saturated, not to provide extra features for the marketing department.
5) Further into the future, 5G is likely to displace wi-fi and broadcast signals, as more devices (TVs, Radios, Consoles, IoT, smart speakers etc) are shipped with embedded 5G. It is quite feasible for example that a 5G smart TV could be rigged up to an existing TV antenna and get oodles of mm wave bandwidth.
6) Wi-fi is saturated in cities and home 5G picocells connected to home broadband are a likely evolution. There's also a particular consumer advantage if new connected devices can operate "out of the box" without the need for local wi-fi configuration.
What do you think happens to the old 2G/3G/4G infrastructure when it's replaced in places like the UK? Seriously, it goes to places like Africa and India.
Smartphones are massive in areas, especially in Africa, without any traditional broadcast or telecoms infrastructure. These people have never had a TV; a smartphone is a revolution for them. No need to develop anything else.
An intent to get cloud providers to share back mods and enhancements is positive. However, an intent to strongarm cloud providers or anyone else into coughing up cash is a disingenuous use of OSS. If you want to sell software, keep it closed source proprietary and sell it; OSS should not be used as a gimmick for financial gain, however big and wealthy the targets are. It corrupts the whole concept.
"Picking up" but a long way from the radical impact they need to have to get profitable. Trouble is, the market is incredibly competitive, apps are the differentiator between smartphone operating systems and app developers aren't seeing the value in developing a third version of their apps for a platform with under 10% market share. So MS have to discount heavily to shift the product and in the process lose all their margins.
WinPho is becoming the betamax of smartphones.
Age is defined as a "protected characteristic" like race or gender. So if you have a real case against a prospective employer, pursue it.
The market is really hot for the right skills - but everyone hiring is doing rigorous testing to make sure you have the skills they need. I suspect the skills mismatch, rather than age, is the real reason these guys aren't getting hired.
Are companies more interested in skills than experience? Yes they are. This is technology, things are developing rapidly, you can have oodles of experience but out of date skills very easily.
Are they discriminating purely on age? I really doubt it, but if they are, take 'em to court over it.
Every time someone alludes that IP is a workable approach for wide area wireless broadcast, I find myself thinking it through and reaching the conclusion that it will never quite get there. The supposition that new technology will *always* be able to leverage more data transit out of existing spectrum has to be false. At some point we will reach critical mass and the spectrum will be saturated.
Therefore it makes sense to use a wide area broadcast technology for data-intensive applications where large numbers of mobile end points moving within a reasonable geographical area wish to receive the same data at the same time. Otherwise we're wasting away spectrum sending the same thing over and over simultaneously. IP multicast doesn't solve this because we're not talking about layer 3 networking, we're talking about mobile/cell layer comms protocols which are explicitly designed around single handset-to-cellsite channels.
IP works for TV services because most viewing is done at home, where wired broadband and local area wi-fi make the bandwidth issue less of a problem.
Is DAB/DAB+/DMB the "right" solution? I don't know; it's true that original DAB is a bit crap, but if maybe you ditch the audio codec - and almost all chipsets, and most sets, being manufactured today also include DAB+ and DMB support. DMB is big in Korea for TV, which gives it some momentum. With the right TX network, mobile reception is better optimised than anything else, including the closest rival, DVB-H.
My main problem with the current status quo is not the underlying technology itself, but rather that its controlled by the radio industry and regulated as radio (audio broadcasting) spectrum. If we go back to what the point is, technically, it's -
"...data-intensive applications where large numbers of mobile end points moving within a reasonable geographical area wish to receive the same data at the same time"
- how much radio industry content really fits that criteria? Live sport, news, stock prices, public events - yes. Automated repetitive middle of the road music playlists with DJs reading liner cards and wall to wall ads, no. A commercial radio station would probably be more bandwidth efficient by letting receivers cache the playlist and ads locally and sending the links - if they're even needed to provide a semblance of live radio - over IP.
So if you leave the radio crowd, and the audio codec, out of the picture, DAB/+/DMB actually gives you the basis of a "Broadcast IP" technology which could feed everyone real time sports, news, weather and finance content, either as audio, video, text and pictures or just metadata to feed apps.
Maybe this isn't quite so bonkers after all.
