|
Jabor posted:Just to give people a sense of scale - 5 nines means you're allowed ~5 minutes of downtime per year. Oh and you have to make sure your mechanisms for failing over don’t cause new and exciting failures that are even harder to work out of.
|
#
?
Oct 9, 2018 03:58
|
|
|
|
| # ? Apr 29, 2024 11:55 |
|
|
VostokProgram posted:Some software systems are designed for 9 9s uptime In what context? I don't think it's possible to measure uptime at that level over reasonable time scales--9 9s would be 1s of downtime every 30 years.
|
|
#
?
Oct 9, 2018 04:19
|
|
|
greatZebu posted:In what context? I don't think it's possible to measure uptime at that level over reasonable time scales--9 9s would be 1s of downtime every 30 years. If there's a bug in these forums and they go down for an hour you don't get to poo poo post for an hour. If there's a bug in medical hardware and it goes down for an hour somebody loving dies. Medical software has pretty insane standards.
|
|
#
?
Oct 9, 2018 04:23
|
|
|
Our CTO once claimed we had nine 9's reliability. Later that year, we blew our nine 9's budget for the next few thousand years.
|
|
#
?
Oct 9, 2018 04:43
|
|
|
ToxicSlurpee posted:If there's a bug in these forums and they go down for an hour you don't get to poo poo post for an hour. If there's a bug in a 9 9s system and it goes down for an hour, it just blew its error budget for the next 100,000 years. If it goes down for one tenth of one second, it just blew its error budget for the next year. I don't think there's any system in existence that can be reliably shown with confidence to be 9 9s and definitely not 8. But if there is one, I would love to read about how they handle their monitoring.
|
|
#
?
Oct 9, 2018 04:47
|
|
|
greatZebu posted:If there's a bug in a 9 9s system and it goes down for an hour, it just blew its error budget for the next 100,000 years. If it goes down for one tenth of one second, it just blew its error budget for the next year. I don't think there's any system in existence that can be reliably shown with confidence to be 9 9s and definitely not 8. But if there is one, I would love to read about how they handle their monitoring. I'm going to guess that you don't know how software development works; there's actually no such thing as an error budget, really. All competently-made software is tested to at least some degree. Automated testing is increasingly a thing and there are places other than "literally attached to a person" that medical equipment can be tested. This is especially true of the software; you can, in fact, simulate way more time than you'd think. So if you simulate, say, a year of use and only get a few seconds of errors then you've shown that the software is solid as gently caress. A medical machine may very well not actually be used for multiple years so that's a pretty significant error rate in the territory of "probably negligible." I have no idea if there's a 9 9s system out there that's been proven; that level of stuff is more typically on the level of "this is our impossible goal, get as close as you can." Rigorous software testing can indicate stuff like "over 2,000,000 runs this chunk failed 6 times." That's not 9 9s but still an absurdly good error rate and good enough for most things. I don't work in med tech but from what I've heard the software practices there are insane for specifically the reason that they don't want software to kill somebody again. The cynical view is they do it so they don't get sued for millions of dollars but, eh...still.
|
|
#
?
Oct 9, 2018 05:14
|
|
|
greatZebu posted:If there's a bug in a 9 9s system and it goes down for an hour, it just blew its error budget for the next 100,000 years. If it goes down for one tenth of one second, it just blew its error budget for the next year. I don't think there's any system in existence that can be reliably shown with confidence to be 9 9s and definitely not 8. But if there is one, I would love to read about how they handle their monitoring. The only time I can remember coming across the "nine nines" statement is for Ericsson's AXD301 switch so here's Joe Armstrong's thesis if you really want. The case study is in chapter 8. I can't speak for how accurate their measurements are but it's by all accounts a pretty reliable switch. In general, I think there are engineering systems in reality that have extremely high reliability. You can find situations where one second of reactor containment failure is probably one second too much, for instance. They're just typically not IT systems.
|
|
#
?
Oct 9, 2018 05:15
|
|
|
ToxicSlurpee posted:I'm going to guess that you don't know how software development works; there's actually no such thing as an error budget, really. This is true for devices too; they are simulated out for years of testing. And yeah, we do the testing both because we want a quality product that helps patients and because we are literally required to by regulatory bodies if we want to sell anything. Keep in mind that this stuff is usually put through its paces for years before we do a formal test.
|
|
#
?
Oct 9, 2018 12:32
|
|
|
I know we've gotten off the subject a bit, but this discussion has been fascinating. For the Mars satellites/robots they could not have any bacteria clinging to them at all, so the testing was as rigorous as was humanly/AI possible. JPL later transferred that technology to a biotech company.
|
|
#
?
Oct 9, 2018 13:59
|
|
|
Crazy thing is playing a ton of factorio actually made me a better programmer.
|
|
#
?
Oct 9, 2018 14:12
|
|
|
Mayveena posted:I know we've gotten off the subject a bit, but this discussion has been fascinating. For the Mars satellites/robots they could not have any bacteria clinging to them at all, so the testing was as rigorous as was humanly/AI possible. JPL later transferred that technology to a biotech company. Nine 9's of reliability* for extra-planetary missions *Does not apply to feet/meters conversion
|
|
#
?
Oct 9, 2018 14:59
|
|
|
Cocoa Crispies posted:Oh and you have to make sure your mechanisms for failing over don't cause new and exciting failures that are even harder to work out of. It never happens during testing, even when we simulate a power outage by cutting the mains feed rather than running the test program, and real outages are rare enough that we don't really have enough information to go on.
|
|
#
?
Oct 9, 2018 19:10
|
|
|
Collateral Damage posted:We have a fun one like that at work. Our backup generator will cut in and out perfectly whenever we test it, but twice during actual power outages we've had issues when power returns. The switchgear disconnects the generator and reconnects the grid, but then the back feed protection seems to think the generator is still connected and disconnects the grid again. Then 10 minutes later the UPS batteries run down and the building goes dark. Yeah, it's frightfully common for stuff to fail like this in distributed computer systems too: https://blog.github.com/2012-12-26-downtime-last-saturday/ https://status.aws.amazon.com/s3-20080720.html https://aws.amazon.com/message/65648/ Factorio's pretty explicitly hitting the same notes as any other complex technological design, which is why it's so addictive for engineers: everything works the way it works, nothing works quite the way you think it does, and most of the problems are your fault.
|
|
#
?
Oct 9, 2018 20:20
|
|
|
ToxicSlurpee posted:I'm going to guess that you don't know how software development works; there's actually no such thing as an error budget, really. There definitely is! At least in some corners of the world, see for example https://www.youtube.com/watch?v=iOoxtpVBQ4I and https://www.youtube.com/watch?v=y2ILKr8kCJU. This conversation is sort of an interesting window into how different groups think about risk and failure; I'm mostly familiar with it from a data processing systems perspective, where uptime/availability is a pretty reasonable metric and there are a bunch of reasonable ways of dealing with transient failures and tradeoffs between cost, reliability, and velocity are very common. But I have also worked on some scientific simulation systems, where "what are the odds that this months-long simulation succeeds" is a much more useful question than "what fraction of the time is this thing working correctly". I think they share some characteristics with high-reliability hardware, where mean time between failures is really what you want to be thinking about--any failure is a really big problem, so it's mostly about minimizing the risk of encountering an error in some relevant timeframe rather than minimizing time spent in an error state per se. To bring things back around to Factorio, the kinds of redundancy and failure tolerance that are available when building your factory seem closer to the style of high-availability distributed systems than high-reliability hardware--create a few layers of redundant backups and some kind of alerting so that if you're losing redundancy there's a person who has been informed and is on the hook for resolving the situation. Then hopefully the human in charge can straighten things out before everything goes to hell.
|
|
#
?
Oct 10, 2018 10:05
|
|
|
Is there any sort of ETA for the next big version?
|
|
#
?
Oct 10, 2018 11:09
|
|
|
Sillybones posted:Is there any sort of ETA for the next big version? They have not announced one on the developer blog. Looks like they are still playing with ideas so I don’t think it will be soon.
|
|
#
?
Oct 10, 2018 13:46
|
|
|
I'd love it if someone actually used Factorio as an interview tool - it hits a lot of concepts present in software engineering and other fields too (as the above discussions show). Separate Q : is there any kind of resource/write-up comparing the UPS costs of different Factorio techniques? E.g. mixed-belts vs. pure-belts, filter-inserters vs. filter-splitters, barrelled-fluid-belts vs. pipes, this sort of thing? I keep telling myself whatever difference is there is going to be negligible, but I still wonder.
|
|
#
?
Oct 10, 2018 18:31
|
|
|
FnF posted:Separate Q : is there any kind of resource/write-up comparing the UPS costs of different Factorio techniques? E.g. mixed-belts vs. pure-belts, filter-inserters vs. filter-splitters, barrelled-fluid-belts vs. pipes, this sort of thing? I keep telling myself whatever difference is there is going to be negligible, but I still wonder. I know I've seen discussion of this sort of thing in the context of megabases. I wasn't specifically looking for it, so what I saw was kinda broader strokes (e.g. bots vs. belts), but with this game, I'll bet someone's done the math on the fiddlier stuff.
|
|
#
?
Oct 10, 2018 18:39
|
|
|
A bot base with no belts or pipes is the most UPS friendly. Any writeup is probably gonna be out of date, seeing as they changed how the belts/splitters/inserters/trains work multiple times. I'll direct you to their blog https://www.factorio.com/blog/ If you're really interested, go visit their forums. Plenty good spergynerd stuff going on there.
|
|
#
?
Oct 10, 2018 19:25
|
|
|
I really want to play this game more but I have such a hard time figuring out how to design my factory and keep it efficient in the process. I basically make it to the point where I’m starting up oil production. I got a bus going (though I have no idea how it works) and haven’t got an idea how to rebalance the belts. I got the first two colours of tech down pat and once oil gets thrown into the mix I get discouraged and find it challenging to move further. Also don’t have a clue how to use robots and railways. I also handcraft more than I should. Also I got a snaking belt like from different resource patches trying to feed my furnaces and my power plants. It all looks lovely. Kraftwerk fucked around with this message at 19:34 on Oct 10, 2018 |
|
#
?
Oct 10, 2018 19:27
|
|
|
Don't worry about efficiency, don't be afraid to tear up a huge chunk of your factory and try again, be organic and handcraft everything until you get into mega factory territory. You'll develop the skills over time and efficiency will come along with it. If you open the game and don't know what to do, find some part of your factory that annoys you and tinker until it fixes. I didn't even touch robots for my first several playthroughs, they're cool, but not necessary to have fun. Trains are a little more important but you can get by without those too if you're determined.
|
|
#
?
Oct 10, 2018 19:34
|
|
|
You'll find good tutorials online but here are some things that weren't immediately clear to me: Build pumpjacks over oil puddles. Raw oil goes into refineries which breaks it down into three products. Everything oil related is further processed in chemical plants. All three products can be made into solid fuel cubes you can then burn in anything that burns coal. Of these, the immediately vital one for science and robots is the petroleum gas. You'll need oil products to build robots and roboports, so don't worry about that just yet.
|
|
#
?
Oct 10, 2018 19:47
|
|
|
Sillybones posted:Is there any sort of ETA for the next big version? I get the feeling the next version is going to be 1.0, so I think they're really trying to polish everything off ready for a big release.
|
|
#
?
Oct 10, 2018 20:28
|
|
|
redleader posted:I get the feeling the next version is going to be 1.0, so I think they're really trying to polish everything off ready for a big release. Nah, they're still talking about 0.17 in all the friday facts. 1.0 is supposed to be a cleaned up version of 0.17 sometime later.
|
|
#
?
Oct 10, 2018 20:35
|
|
|
Yeah the reason 0.17 is taking so long is because it's going to be the last pre-1.0 version, so they're trying to cram in all of the big changes and ideas.
|
|
#
?
Oct 10, 2018 20:54
|
|
|
Rewriting the entire gui framework probably took a fair bit of the 0.17 time by itself. I can't wait for this update, looks like a lot of good QOL stuff coming.
|
|
#
?
Oct 10, 2018 20:55
|
|
|
The most important thing about oil processing is to do the 3 different pipe lengths for outputs from the refinery. Crude gets 1 pipe segment out the back and then joins up left/right. Light oil gets 2 straight pipes out then left/right. Heavy gets 3 pipes out then left/right. Once you get that, you can have a row of many refineries, and they all pump out into easy to manage pipes, so you can tile them together. When you first start out doing oil, just set down a crap ton of storage tanks down the line, and thanks to the pipe separating above, it's easy to get going. If you run out of room for your oil, just make more tanks. Or just deconstruct the tanks and dump it out.
|
|
#
?
Oct 10, 2018 21:05
|
|
|
captain innocuous posted:The most important thing about oil processing is to do the 3 different pipe lengths for outputs from the refinery. I made a quick video tutorial about this a while back, sounds like it might be useful to a few people again https://www.youtube.com/watch?v=kk_UdBr5G8Y
|
|
#
?
Oct 10, 2018 22:49
|
|
|
captain innocuous posted:The most important thing about oil processing is to do the 3 different pipe lengths for outputs from the refinery. Before you do this, get VERY comfortable with how pipes connect/don't connect when working with undergrounds and things this close to each other. All too often I accidentally drop a pipe without realizing it's going to connect to something, and then I gently caress up all of them with some incorrect liquid.
|
|
#
?
Oct 10, 2018 23:40
|
|
|
Always user underground pipe except on turns. It's more efficient for long length. When it's not more efficient it keeps your streams from crossing anyway. Lot easier to avoid mix-ups if the only place they can join is at corners.
|
|
#
?
Oct 11, 2018 00:31
|
|
|
zedprime posted:Always user underground pipe except on turns. It's more efficient for long length. When it's not more efficient it keeps your streams from crossing anyway. Lot easier to avoid mix-ups if the only place they can join is at corners. And the biggest bonus, you can get through your pipe maze without having to run all the way around long sections.
|
|
#
?
Oct 11, 2018 01:13
|
|
|
Also if you delete a full storage tank, all that fluid just disappears and you lose it forever.
|
|
#
?
Oct 11, 2018 01:19
|
|
|
This is not necessarily a bad thing, either!
|
|
#
?
Oct 11, 2018 03:31
|
|
|
Removing a full tank should spill the contents, counting as a big pollution hit. Speaking of pollution, those times I've complained about insufficient attacks? Well my wish came true. I took this screenshot a few hours ago, detailing the novel experience of having enemies launch concerted attacks on me. My response was to beef up my single turrets with lines of them. I connect power and lighting, with the idea of adding walls if it became necessary.  Well, I spent the rest of those hours fighting off wave after wave of the bastards. My bullet production was barely automated at all when this started, so I had to keep running back and forth with small batches of bullets to feed the turrets that were killing the swarms. I also got several attacks on undefended parts of my base, including my power grid and science research. There was a... nest? nearby my base, so I launched an attack on it. They easily repelled it-- those loving worm things outrange both me and my turrets. I built a heavy armour suit which helped me not die instantly, and started to build turrets and wall segments closer and closer to the nest. My best method so far was to have a line of turrets by the nest, build turrets for the worms to shoot at, then run in and throw grenades. This didn't work very well but it was the best I had. What sort of weapons should I be looking at here? I'm pretty low tech still, I've got red pcbs in steady production but no batteries yet. I need mil science before I can research flame throwers. If the aliens level up I'm seriously hosed, I can barely hold them off as it is. Count Roland fucked around with this message at 04:04 on Oct 11, 2018 |
|
#
?
Oct 11, 2018 04:02
|
|
|
Count Roland posted:What sort of weapons should I be looking at here? I'm pretty low tech still, I've got red pcbs in steady production but no batteries yet. I need mil science before I can research flame throwers. Red bullets for everything, plus rockets so you can outrange the worms. Getting military science up should be a priority. I do it before I even set up oil.
|
|
#
?
Oct 11, 2018 04:06
|
|
|
Military science upgrades help a tremendous amount. Bullet damage/speed and grenade damage are amazing help early on. Once you get a car you can pretty easily knock out bases.
|
|
#
?
Oct 11, 2018 04:09
|
|
|
TasogareNoKagi posted:Red bullets for everything, plus rockets so you can outrange the worms. Getting military science up should be a priority. I do it before I even set up oil. Rockets eh, no I definitely don't have those. I'll rush for mil science and pray I have enough time. For the time being I'll put armour piercing bullets and grenades on major production, as I literally have nothing better.
|
|
#
?
Oct 11, 2018 11:44
|
|
|
Are you close to the tank? The tank is v. good in midgame. Ignore its main cannon; just use its machine gun with red ammo. It shreds bases and biters real good.
|
|
#
?
Oct 11, 2018 11:51
|
|
|
necrotic posted:Military science upgrades help a tremendous amount. Bullet damage/speed and grenade damage are amazing help early on. Once you get a car you can pretty easily knock out bases. Also, the turret damage and bullet damage upgrades stack with each other, so you're going to get more damage out of a stack of bullets by putting it into turrets. Piercing bullets with all the damage upgrades through blue level will tear through everything up to and including big biters. Although you need oil for the components, you can get modular armor and some armor modules with just RGM tech. A 5x5 modular armor with 2 batteries, 2 shields, and 13 solar panels should keep itself charged most of the time (just run back to base at night) and is enough to tank a couple worm hits until you can plop turrets down. You'll need to make acid to make batteries, but you shouldn't need too much to make the handful of batteries needed for this armor. Early oil tech just drop down 1 plant making sulfur, 1 making acid, and leave it running into a tank - this will give you enough acid to make a good deal of batteries and get you started on blue chips later down the line. I like to put a buffer chest between the output of my grenades, red bullets, and turrets and the military science, so I can usually pick up a couple stacks of everything before going out to clear things. Military science is the one thing that doesn't need to run continuously so I usually get by with just a couple assemblers and alternate between military and non-military techs. The science belts hold enough of a buffer that production rarely backs up. Military tech is really heavy on the iron usage, but it can also be the first time you face serious issues with copper. Turrets and red bullets both take decent chunks of it. A tip for clearing with turret creep: Put bullets and turrets on your hotbar and filter the slot with middle-click. I personally use slots 3+4 for this (on the second row, after getting toolbelt research) so dropping a line of turrets is just pressing 3, dragging a bit, pressing 4, ctrl-right clicking down the line, and the turrets are up and shooting within seconds. You want a lot of turrets and bullets in your inventory, at least 1 stack of turrets and 10 stacks of ammo. Yellow ammo should be fine with small and medium biters, but you want to get red ammo up ASAP. A low-tech modular armor helps a ton with turret creep since you can drop the turrets in much closer. My personal weapons of choice are the shotgun and rocket launcher (with explosive rockets), and I usually stick my machine gun and its ammo in my inventory when clearing bases. Even with just the shotgun and a few damage/speed upgrades, you can drop a line of turrets to take out any spawned biters (and creep on worms if necessary) then rush in with your shotgun and just blast down the spawners. When I get the tech up, I use explosive rockets pretty much exclusively (plus shotgun for clearing out trees). With the max tech upgrades for RGBM, they two-shot any big biters in their explosion range, and spamming them on bases keeps the spawns down. Your biggest priority when clearing should be taking out the spawners; if you just kill the biters the spawners will replace them really quickly, and often with tougher enemies since evolution only kicks in on newly-spawned biters. Explosive rockets do their job perfectly in the midgame, both taking out spawners quickly and weakening or killing the already-spawned biters.
|
|
#
?
Oct 11, 2018 15:54
|
|
|
|
| # ? Apr 29, 2024 11:55 |
|
|
Count Roland posted:I took this screenshot a few hours ago, detailing the novel experience of having enemies launch concerted attacks on me. My response was to beef up my single turrets with lines of them. I connect power and lighting, with the idea of adding walls if it became necessary. Count Roland posted:There was a... nest? nearby my base, so I launched an attack on it. They easily repelled it-- those loving worm things outrange both me and my turrets. I built a heavy armour suit which helped me not die instantly, and started to build turrets and wall segments closer and closer to the nest. My best method so far was to have a line of turrets by the nest, build turrets for the worms to shoot at, then run in and throw grenades. This didn't work very well but it was the best I had. You can get by with just yellow bullets for a long time, so I'd suggest not worrying about upgrading to red bullets until you've cleared out the biter nests under your pollution cloud. Red bullets also take a loooooooooot more resources per bullet than yellow does.
|
|
#
?
Oct 11, 2018 16:09
|
|
