|
Bulgakov posted:that seems like a less bad idea than other possible ideas pushed down from on high, maybe yeah, i mean, in principle forcing projects to use a common framework is much better than letting them all do their own wacky thing. anything you do to improve that framework — hacking on it, building libraries on top of it, figuring out how to configure it better, learning how to use everything it gives you — helps the whole org at once. of course it would’ve been better if they’d had the vision to do that before all this code got written, and for all i know this specific framework is total poo poo, but at least in theory it’s still a better-late-than-never sort of move
|
# ? Nov 9, 2019 05:19 |
|
|
# ? Jun 3, 2024 20:35 |
|
jit bull transpile posted:my org is currently in the process of overreacting to this by pushing hundreds of disparate groups into a single shared platform that will require 90% of us to rewrite our code from the ground up because they refuse to support frameworks used all over the place in favor of their one true God Spark. every team doing its own thing poorly -> centralize -> no funding for continuous improvement to centralized infrastructure -> the centralized infrastructure is outdated and holding everyone back -> teams abandon centralized infrastructure to do their own thing
|
# ? Nov 9, 2019 05:23 |
|
Fiedler posted:every team doing its own thing poorly -> centralize -> no funding for continuous improvement to centralized infrastructure -> the centralized infrastructure is outdated and holding everyone back -> teams abandon centralized infrastructure to do their own thing an ancient pendulum
|
# ? Nov 9, 2019 05:23 |
|
abigserve posted:- colos are loving expensive. 25k/year per rack not including power costs. Keeping in mind you need two for redundancy... 25k a year is like 10 x1e 4xlarge instances. its nothing. the compute/memory density in that 25k a year rack could be like 50x more than the same cost in aws vms also lol once theyre racked and cabled thats pretty much it dude. most dcs are lights out unless theres remote hands doing something. you make it sound like some datacenter monkey has to be there turning gears and adjusting the steam furnaces 24/7 like metropolis
|
# ? Nov 9, 2019 06:40 |
|
we've vertically integrated our internal devops pipeline so programmers can wander freely amongst the racks
|
# ? Nov 9, 2019 06:49 |
|
rjmccall posted:theres a poo poo ton of stuff about running your own servers that scales extremely poorly, especially because its often individual teams within a company that organically develop their own server needs, and the sort of org that just says yeah, you need your own server usually follows it up by saying and you should run it yourself instead of bothering the extremely important team of masturbating domain admins, and suddenly you look up and Web Initiative Development Team #3 is putting 40% of some poor saps time towards swapping hard drives on the local build cloud and investigating backup solutions instead of doing the job they were actually hired to do thats funny ive experienced the exact opposite. most of the time the people doling out hardware is some ultra entrenched graybeard gatekeeper cult
|
# ? Nov 9, 2019 06:52 |
|
i work for an amazon subsidiary so we just use AWS for half cost and that seems fine
|
# ? Nov 9, 2019 06:58 |
|
pram posted:25k a year is like 10 x1e 4xlarge instances. its nothing. the compute/memory density in that 25k a year rack could be like 50x more than the same cost in aws vms it basically is that though, even if you aren't physically there all the time your whole life is; - hardware/software (aka hypervisor, array, ups, ilom, etc.) upgrades - hardware replacements - firmware bug hunting - network infrastructure bitch work yeah in a perfect world where every data centre is rows of perfect current gen 1ru pizzaboxes you're right, but back on planet earth hardware goes out of date, fails, or never works right to begin with, and some poor fuckers aka the infrastructure team have to deal with it while bean counters sit at the top going heh look at all this money we're saving Notorious b.s.d. posted:??? if you compute the cost of those people running your infrastructure versus the savings of not sending it to the cloud you get what i'm saying
|
# ? Nov 9, 2019 07:08 |
|
pram posted:thats funny ive experienced the exact opposite. most of the time the people doling out hardware is some ultra entrenched graybeard gatekeeper cult doling out standard hardware to individual employees, sure. what i'm talking about is when your team needs, like, a wiki or a buildbot or something, and usually this means you have two options. the first is, maybe the it group has a standard solution for your problem, but it's probably some insanely locked down out-of-date atlassian product that the it group is just totally disinterested in supporting. and the second is, you can go gently caress yourself and (maybe after a political fight) take care of everything within your group, which is a path that will eventually lead to you maintaining your own server room in the office next to gary because gary likes having the ac really low anyway
|
# ? Nov 9, 2019 07:31 |
|
rjmccall posted:yeah, i mean, in principle forcing projects to use a common framework is much better than letting them all do their own wacky thing. anything you do to improve that framework — hacking on it, building libraries on top of it, figuring out how to configure it better, learning how to use everything it gives you — helps the whole org at once. of course it would’ve been better if they’d had the vision to do that before all this code got written, and for all i know this specific framework is total poo poo, but at least in theory it’s still a better-late-than-never sort of move yeah the problem here is spark is loving garbage and nearly everyone in my group (5000+ people) use map reduce which they're just dropping support for. this isn't just a simple framework swap, people will basically have to rewrite everything from first principles.
|
# ? Nov 9, 2019 08:01 |
|
jit bull transpile posted:yeah the problem here is spark is loving garbage and nearly everyone in my group (5000+ people) use map reduce which they're just dropping support for. this isn't just a simple framework swap, people will basically have to rewrite everything from first principles. map reduce has been dead for years and spark is genuinely better in every way you should have started that migration five plus years ago. if not to spark then to something else I’m sure it sucks to try and catch up on years of work un-done but MR has problems and isn’t actively developed
|
# ? Nov 9, 2019 08:30 |
|
abigserve posted:it basically is that though, even if you aren't physically there all the time your whole life is; your cloud provider has all the same costs you do, and they are offering short term contracts, even hour to hour. they will want to be paid for this service
|
# ? Nov 9, 2019 08:31 |
|
Notorious b.s.d. posted:map reduce has been dead for years and spark is genuinely better in every way spark is loving terrible, makes it nearly impossible to find logs of any kind of failure, swallows stacktraces constantly, and is a never ending quest to find the exact memory settings to chant that will make it not crash arbitrarily. it might be the new hotness but it is not ready for primetime in any way and you will pry crunch running on mr out of my cold dead hands.
|
# ? Nov 9, 2019 09:09 |
|
Notorious b.s.d. posted:your cloud provider has all the same costs you do, and they are offering short term contracts, even hour to hour. there is no way you could offer even 1% of the services azure, aws or even gcp could provide with any form of reliability or uptime in a bespoke on-prem environment for even a cost approaching close to what they will charge you unless you are dealing with atypical compute/storage requirements (like if you ran a supercomputer facility or were dealing with exabytes of storage) the reason I'm arguing with you about this is because I just got out of an environment where the higher ups had the same exact line of thought you do now five years ago and guess what? They couldn't provide the service, the clients got pissed off in a previously extremely well regarded central IT department and the org ended up doing a force shift to the cloud to try to re-onboard all the cloud-based shadow it.
|
# ? Nov 9, 2019 09:14 |
|
rjmccall posted:doling out standard hardware to individual employees, sure. what i'm talking about is when your team needs, like, a wiki or a buildbot or something, and usually this means you have two options. the first is, maybe the it group has a standard solution for your problem, but it's probably some insanely locked down out-of-date atlassian product that the it group is just totally disinterested in supporting. and the second is, you can go gently caress yourself and (maybe after a political fight) take care of everything within your group, which is a path that will eventually lead to you maintaining your own server room in the office next to gary because gary likes having the ac really low anyway no you would just use azure devops, obviously
|
# ? Nov 9, 2019 09:39 |
|
Notorious b.s.d. posted:your cloud provider has all the same costs you do, and they are offering short term contracts, even hour to hour. maybe the same costs for iaas (probably not, though, because they get discounts every step of the way that your company won't get, because your company isn't, for example, the #1 purchaser of intel cpus in the world). but if you want to run software (e.g. a database) on that infrastructure, that's a different cost model entirely.
|
# ? Nov 9, 2019 09:58 |
|
CRIP EATIN BREAD posted:yeah gently caress running a on-premise datacenter. it's not the physical hardware thats an issue, it's having redundant, highly available network connectivity through geographically diverse physical locations integrated in a unified mechanism. just get your own as, get transit with 2 different carriers. you need a couple of routers for the edge (take a default route not a full table) and some cheap dumb switches that can do very basic mpls for the core. duplicate in two different dataceters. it's really not so bad imo
|
# ? Nov 9, 2019 10:20 |
|
pram posted:25k a year is like 10 x1e 4xlarge instances. its nothing. the compute/memory density in that 25k a year rack could be like 50x more than the same cost in aws vms people will pay the 50x more just so they don't have to go replace the odd dead drive twice a year apparently
|
# ? Nov 9, 2019 10:27 |
|
Notorious b.s.d. posted:map reduce has been dead for years and spark is genuinely better in every way working things are working things, defensively rewriting your poo poo from the ground up to conform to some idea of what is current is an insane idea. at minimum if the change is going to amount to a rewrite anyway, which this clearly will, it makes sense to delay it as long as is practical.
|
# ? Nov 9, 2019 10:28 |
|
spark isnt great but a lot of those exact complaints apply to mr/yarn lol
|
# ? Nov 9, 2019 11:18 |
|
my stepdads beer posted:people will pay the 50x more just so they don't have to go replace the odd dead drive twice a year apparently yeah but you get to fire a bunch of annoying as gently caress CJs so its worth it.
|
# ? Nov 9, 2019 13:09 |
|
CJs like to pull out that “the cloud is just someone else’s computer” line like it protects them. that’s the entire point, your business wants someone else’s computer so they don’t have to deal with your poo poo anymore.
|
# ? Nov 9, 2019 13:11 |
|
As someone who has worked for a VAR I'd go cloud just to not have to deal with VAR sales
|
# ? Nov 9, 2019 13:28 |
|
abigserve posted:there is no way you could offer even 1% of the services azure, aws or even gcp could provide with any form of reliability or uptime in a bespoke on-prem environment for even a cost approaching close to what they will charge you i mean, i live in that world now reliability and uptime in your average datacenter provider are better than what you get with AZ failover in aws, my friend not to mention, i work for a big enough company that they have insane services available to developers based on in-house k8s and poo poo it's ... not awful. it does not have as good customer service, though. sure, you can ring a guy at his desk, but that doesn't mean he will be nice to you! abigserve posted:unless you are dealing with atypical compute/storage requirements (like if you ran a supercomputer facility or were dealing with exabytes of storage) i don't doubt that work has exabytes of storage online but not that much of it is in "modern" platform-as-a-service scenarios it takes a long, long time to turn a big ship abigserve posted:the reason I'm arguing with you about this is because I just got out of an environment where the higher ups had the same exact line of thought you do now five years ago and guess what? They couldn't provide the service, the clients got pissed off in a previously extremely well regarded central IT department and the org ended up doing a force shift to the cloud to try to re-onboard all the cloud-based shadow it. the smaller you are, the less the 90%+ margins on cloud poo poo matter especially if you can sign long term contracts with the cloud vendors, in which case the 90% margins go down to like 50%
|
# ? Nov 11, 2019 05:26 |
|
Trimson Grondag 3 posted:CJs like to pull out that “the cloud is just someone else’s computer” line like it protects them. that’s the entire point, your business wants someone else’s computer so they don’t have to deal with your poo poo anymore. it's not a line it's the beginning of a conversation what are you paying for, and why? you would be surprised by how much better people feel about their gigantic aws bills when they start to consider alternative scenarios, and plan out the costs based on varying expansion needs (and urgencies!)
|
# ? Nov 11, 2019 05:27 |
|
Has anybody run a quote for extended 2008r2 support? The only number I could see kicking around is $44/yr for a 2-core license, no idea if that is in the right ballpark or not.
|
# ? Nov 11, 2019 18:39 |
|
just double it and then if management still doesn't want to upgrade atleast you have a nice new bonus for dealing with it.
|
# ? Nov 11, 2019 19:43 |
|
Notorious b.s.d. posted:i work for a big enough company that they have insane services available to developers based on in-house k8s and poo poo "how expensive could it really be to replicate the services offered in aws and azure," the cto asserted confidently.
|
# ? Nov 11, 2019 20:33 |
|
Fiedler posted:"how expensive could it really be to replicate the services offered in aws and azure," the cto asserted confidently. not that expensive, except for the part where aws and azure are happy to add half a datacenter's worth of capacity with like, ten minutes of notice the services part isn't the hard part
|
# ? Nov 11, 2019 20:36 |
|
lol sure. buildings full of 6.5 figgie developers working for years creating various database services. the same for paas compute offerings. minor trivialities.
|
# ? Nov 11, 2019 20:55 |
|
standing up a k8s cluster and cutting big checks to oracle isn't even close
|
# ? Nov 11, 2019 20:58 |
|
surely openstack will allow me to keep pace with the hyperscalers
|
# ? Nov 11, 2019 22:20 |
|
Trimson Grondag 3 posted:surely openstack will allow me to keep pace with the hyperscalers remember when people actually believed this the only sadder idiots were the folks trying to sell openstack
|
# ? Nov 11, 2019 23:16 |
|
So is Microsoft going to release an EdgiumBook? Satya has mentioned the company is not about Windows anymore and so it would make sense as competition to ChromeBooks.
|
# ? Nov 12, 2019 17:31 |
|
MrMoo posted:So is Microsoft going to release an EdgiumBook? Satya has mentioned the company is not about Windows anymore and so it would make sense as competition to ChromeBooks. the only added cost of putting windows on them is the profit microsoft otherwise extracts, so why wouldn't it be windows? chromeos had no differentiating functionality.
|
# ? Nov 12, 2019 17:33 |
|
given how many times they return to the "windows that doesn't run windows software" well, a thin client for their web services makes about as much sense as anything
|
# ? Nov 12, 2019 17:44 |
|
at least then the absolute dogshit performance of the arm surface doesn't matter
|
# ? Nov 12, 2019 17:45 |
|
Cybernetic Vermin posted:the only added cost of putting windows on them is the profit microsoft otherwise extracts, so why wouldn't it be windows? chromeos had no differentiating functionality. In ChromeOS only small things tend to break and updates are almost invisible. It’s also a cloud storage client and Satya will love to get more people hooked into Microsoft services. I think it would be a pretty smart move as Microsoft experience on cheap hardware is poo poo, and this would raise the bar.
|
# ? Nov 12, 2019 18:10 |
|
MrMoo posted:So is Microsoft going to release an EdgiumBook? Satya has mentioned the company is not about Windows anymore and so it would make sense as competition to ChromeBooks. make windows 2020 a linux satya apple won't even know what courage is by then
|
# ? Nov 12, 2019 19:13 |
|
|
# ? Jun 3, 2024 20:35 |
|
MrMoo posted:In ChromeOS only small things tend to break and updates are almost invisible. It’s also a cloud storage client and Satya will love to get more people hooked into Microsoft services. haha, windows restricted to just run edge is vastly superior to chromeos, the only thing chromeos has going for it is that it's free. windows is free to microsoft.
|
# ? Nov 12, 2019 19:18 |