
sounds about right
Over the years I've probably had more power outages caused by UPS failures than I have had UPSes saving the DC from a power feed failures.
58 publicly visible posts • joined 20 Nov 2013
Yes, I was going to say that this doesn't seem to be a realistic as well.
Performance scaling isn't close to being linear even in a fully meshed system, and most HPCs are divided up into cells, pods, and racks that you already try to keep your job inside to minimise unacceptable latency. Few jobs are going to be suitable for going distributed-dc compute, certainly not current AI training.
Even Frontier classifies a large job to as one which takes up 1/4 of the machine, iirc.
Sure, but putting a few tens of kilowatts of panels on the roof of a DC isn't going to touch the sides. We're using 50-100kW a rack. Water cooled h200 systems are pushing 250kW a rack, and the next gen open rack standard is looking to double that.
It isn't pointless, but it isn't close to being enough.
We had a laser link that worked fine for months, then started dropping out for an hour each day as the setting sun had moved round to where it was swamping the receiver.
Tried extending the hood and making comedy goggle type things out of pvc tube, but in the end we replaced it with a microwave link.
We trialed immersion cooling years ago in our DC. It just made everthing more difficult. Swapping out a dead server was a horrible task, but it was the non-obvious things that were worse - any cables in the oil turned rigid as the plasticiser was leached out, and the oily coolant wicked up inside any cable. Fun when the leak alarms go off and you find oil dripping down the pdus and switches. Still finding that crap in places. Also with ours being a horizontal 40u-ish tank there wasn't really any overall space saving anyway.
Flew into HIA once just behind a flight from Korea. The immigration system also had the 'open eyes!' thing which took down a gate as the staff couldn't overide it. All the staff thought this was hilarious. You could actually see the supervisor switch to panic mode as the Koreans started reaching the other desks.
That was a particularly long evening in Qatar's already legendary immigration queue.
It's an irrellevant annoyance that Google insists on using in [whatever Google Now Cards Are Called This Week] as far as I can tell.
Amp pages are so slow to load on my mobile that I can normally edit the amp out of the URL before the page has loaded and get to the proper site very quicky. And which inevitably works just fine anyway. And includes all the content as the site designer intended. (not always a good thing.)
Used to infuriate me but then some kind soul made https://addons.mozilla.org/en-GB/firefox/addon/amp2html/ and I can carry on blissfully unaware of amp for a bit longer.
Come on elReg, do the right thing and dump amp instead of just bitching about it.
I'd assume they'll do the same annoying thing they did when they last started hiding bits of the URL such as 'http[s]' which was that when you copied the host bit of the URL to paste into a shell window they'd "helpfully" add the invisible protocol bit back and you'd find yourself trying to, say, ssh into "https://example.com/"... Genius.
Ditched Chrome after a bug report for this crappy protocol and path hiding of essentially 'this is v.stupid and at a minimum please for the love of Dog have a flag to disable it' was closed with wontFix and an arrogant retort from the Chrome devs.
We once had just moved to a temporary site using a particular building for the main servers because that was where the previous occupants had their server room, and it had a genset.
Shortly after moving, one evening I get a call from my guy covering the late shift that there was a power outage, and although miraculously the vintage generator managed to start itself, both UPSes were screaming about not getting a feed. He'd done all he could reasonably do, including manually switching over to the generator feed, but no joy.
When I get on site I'm met by my guy looking a little sheepish and a strong smell of burning. Turns out that whatever the previous inhabitants had considered valuable enough to provide generator power to, the server room wasn't one of them. The single plug at the photocopier station just up the hall, though...
Bless his heart, but he'd strung together 4 or so extension cables and evicted the Xerox.
Red hot cables, an exciting scorch mark all along and up over the hall walls, and a smoking and now eternally dead socket, plus the knowledge that perhaps the fire alarm should be re-tested but by someone competent this time.
It seems properly odd that a company the size of IAG, especially if you get their 20-25% owners Qatar Airways on-board too, would outsource something so critical to their operation and reputation. The licensing bill must surely be fearsome.
But what do I know, my clothing sense is as deeply unfashionable as in-house IS is in this era of perpetually renting some cloudy 3rd party webservice saas bollocks.
Once upon a time in Bahrain yours truely was helping service a large UPS. This included replacing the 4 large shelves of large batteries while the DC was fed through the much smaller backup UPS.
We were just 3/4 of the way through when there was a national power outage. The backup UPS kept the critical systems up, but it would only hold for 15 mins.
Picture 3 guys in close proximity desperately attempting to connect up the last load of huge 48V batteries, on earthed metal shelves, in a tight space, with no AC, no air circulation, a weak emergency light, and +40C outside air temperature. Humid as hell, and with every surface at nice cool DC temperature, heavy condensation was everywhere. On us, running down the walls, and... on the rubber-coated handles of our electrical spanners...
So many shocks.
So many.
Good job guys.
Don't get company values such as 'uprightness', and 'humbleness' at many other places.
I'd left the Big M group years before this, but we were well into a outsourcing and centralisation program, pushing lots jobs to India. Despite it being as unpopular with the users as these things always are, we went full steam ahead on Helldesk and desktop support, but the plans went much further. We were pushed to an expensive US company for HP-UX, Oracle, and Navis for example, which didn't seem particularly smart since Maersk had plenty of domain knowledge and was big enough to support centralised internal teams. Can't say they didn't know their stuff though.
There was also talk of centralising port infrastructure until the speed, cost, and reliability of Internet/IPLCs in many locations was brought up.
Yes the roll-out of fibre is expensive, but chances are once you've got a fibre to each house, that's pretty much all the cabinet/prem network investment you need for the next 20* years (copper has lasted +100 years after all). Amortised over that sort of time frame it really shouldn't affect bills. Unfortunately publicly listed companies rarely withstand shareholder pressure for short term profits, so strategic planning has become something of a rarity in large Western businesses.
Maybe I'm just salty that you can get pukka FTTP up in darkest Malham, yet the current place I am staying at while over here is 10 miles outside Reading, supposedly a UK IT hub, and I have the indescribable pleasure of 4M ADSL that drops packets like a meth dealer high on their own product when it's windy or raining... I'd say it's like going to a 3rd world country, except many of them have better Internet access than this.
* Maybe swap out transceivers now and again, maybe switchout PON for AON, but we're always told running the cable is the actual expensive bit.
'Miniature' one and two way ADSB transceivers are already available which would enable your multirotor to participate in TCAS avoidance, but they are of the $2k region, eat power, and aren't really miniature enough for smaller than 450 class craft. The sort of proximity flying your typical 150-250 class is doing is not going to interfere with aircraft in any case, so it would just be dead weight and something else to break in a crash!
Have found WD enterprise drives to not be especially reliable (5/6 dead in 4 years), and the warrenty not worth the effort - apparently no-one buys drives in one country then moves to a different one.
Not as bad as the Seagates before that, and 4 of them had the decency to fire of smart alerts a short while before dying.
Really hoping they don't fuck up HGST with the 'integration'.
I thought it was standard practise for US government officials to have multiple opaque email accounts and even off-site accounts for the purposes of attempting to avoid the records act?
In my opinion, each government office should have a published address and official business should be directed there rather than to the individual, whatever fake name they're using for their current email.
We also need to crack down on officials claiming not to use email, as first that is ridiculous in this day and age, and secondly a strong pointer that they're up to no good if they're trying to keep stuff off the official record.
Maybe they could introduce two tiers of flights and have them active on alternate flights. One for the fearful where you can only fly with your passport and boarding card, and another where you can fly with what we used to, such as a laptop and a bottle of pop.
Let The Market Decide - that's the Republican way, right?
But that's just it, the speed you can go on a particular bit of road is already variable. The posted limit, which used to be set at the 95th percentile of best case, is the _maximum_.
If it is rainy, or foggy, or kicking out time at the school, you probably should not be at say the 50 limit that stretch of road otherwise warrants. Unfortunately there are many retards and unobservant drivers around who can't or won't drive to the conditions. This leads to situations where we have limits being artificially reduced on broad swathes of road, just to cope with one slightly tight bend near one end, and ultimately to generally slower and slower limits being imposed.
This in turn leads to devaluing the lower limits, driver resentment of said low speed limits, and then more 'speeding' because people drive to the perceived risk of the bit of road they are on.
The better action is to engineer that bit of road so that the speed the majority of people would drive below is safe rather than attempt to force lower limits. If you want engage drivers in slowing themselves down, you need to have road engineers redesign the road so drivers believe it to be riskier. For instance, planting trees down the edge of the road leads to a significant decrease in driver speeds, as do things like pedestrian refuges in the centre of the road.
If you already have thermal power stations, then even after the 2nd or 3rd stage turbines the steam still has heat to help boil off seawater, and as you need to cool the steam off anyway the combined power and water plant can work out cheaper overall than RO.
Combine that with using TSE for agriculture and city planting irrigation.
The nice thing about TSE is that as the city grows, you get more of it. You can see this in action in Oman where the trees along the side of the highways reach further out of town every year.
The ME's municipal planting is almost exclusively watered with TSE, since they have a decent amount of it and Islam apparently has some restrictions against using it for crops (I have been told).
You've had no problems with Win10? Seems unlikely!
Guess you haven't changed your wi-fi PSK yet, or used Edge to go to youtube.com/tv, or tried to select all emails in a folder, or tried dragging a window between monitors without moving it at 1,000kph, or set the system sounds to a lower volume and rebooted? etc., etc. Admittedly, I've hit nothing really show-stopping, but it looks like a perfect example of Agile development done wrong.
Also, I dunno about you, but I'm getting snowblindness from all the white. No themes, no dark colour schemes, no user colour schemes, removal of keyboard shortcuts and most right click actions. It's getting to be as annoying to use as OSX, and even Apple started to use the right button a few years back.
The important question to ask when buying flash-based storage is what happens when the spare area is used up.
Intel rather obnoxiously brick the drive ensuring that data recovery is impossible. Why they wouldn't fail to read-only I don't know. At least that way you could recover any data off the thing.