SoylentNews is people
SoylentNews
https://www.realworldtech.com/forum/?threadid=183440&curpostid=183486
Guys, do you really not understand why x86 took over the server market?
It wasn't just all price. It was literally this "develop at home" issue. Thousands of small companies ended up having random small internal workloads where it was easy to just get a random whitebox PC and run some silly small thing on it yourself. Then as the workload expanded, it became a "real server". And then once that thing expanded, suddenly it made a whole lot of sense to let somebody else manage the hardware and hosting, and the cloud took over.
Do you really not understand? This isn't rocket science. This isn't some made up story. This is literally what happened, and what killed all the RISC vendors, and made x86 be the undisputed king of the hill of servers, to the point where everybody else is just a rounding error. Something that sounded entirely fictional a couple of decades ago.
Without a development platform, ARM in the server space is never going to make it. Trying to sell a 64-bit "hyperscaling" model is idiotic, when you don't have customers and you don't have workloads because you never sold the small cheap box that got the whole market started in the first place.
Submitted via IRC for Bytram
Channeling the late Steve Jobs, Linux kernel king Linus Torvalds this week dismissed cross-platform efforts to support his contention that Arm-compatible processors will never dominate the server market.
Responding to interest in Arm's announcement of its data center-oriented Neoverse N1 and E1 CPU cores on Wednesday, and a jibe about his affinity for native x86 development, Torvalds abandoned his commitment to civil discourse and did his best to dampen enthusiasm for a world of heterogeneous hardware harmony.
"Some people think that 'the cloud' means that the instruction set doesn't matter," Torvalds said in a forum post. "Develop at home, deploy in the cloud. That's bullshit. If you develop on x86, then you're going to want to deploy on x86, because you'll be able to run what you test 'at home' (and by 'at home' I don't mean literally in your home, but in your work environment)."
For Torvalds, this supposedly unavoidable preference for hardware architecture homogeneity means technical types will gladly pay more for x86 cloud hosting, if only for the assurance that software tested in a local environment performs the same way in the data center.
Jobs during his time as Apple's CEO took a similar stance toward native application development, going so far as to ban Adobe's Flash technology on devices running iOS in 2010. For Jobs, cross-platform code represented a competitive threat, bugs, and settling for lowest-common denominator apps.
(Score: 1, Insightful) by Anonymous Coward on Monday February 25 2019, @11:33AM (18 children)
Raspberry Pi is your toy to play around with these days. So not sure about ARM ....
I guess it also means that phones with x86 will never be possible either because ARM dominates that space.
(Score: 5, Interesting) by bobthecimmerian on Monday February 25 2019, @12:01PM (11 children)
The Raspberry Pi and devices like it are fine pieces of hardware, but I'd hate to have it as my primary development environment.
But I think Torvalds is downplaying an important aspect of historical context: when x86 took over the data center, I suspect (but admittedly can't prove) that a lot more data center software workloads were in C++ and C. That made the headaches of cross-compilation significant. Today if you're writing your server side code in PHP, Python, Nodejs, C#, Java, and all of the .NET and JVM derivatives then your developers don't have to worry about cross-compilation. The people working in C++, C, Swift, Go, D, and Rust still care but there are enough of the other group that I think ARM should be able to establish a foothold.
(Score: 5, Interesting) by DannyB on Monday February 25 2019, @03:22PM (8 children)
Thank you.
As a Java developer, I was going to bring up that changing OSes, and even architecture would have very little effect. I could develop on Windows x86 and deploy on Linux on ARM.
There are some things in Java that use native libraries. Even these are packaged for the various OS / architecture combinations. I know they exist. The only one I can think of at the moment is for Serial I/O. Not something the vast majority of Java applications even do.
Back in 2014, my boss's, boss's boss came into my office, handed me a raspberry Pi (model 1 with 512 MB) and said "here, do something cool with this and let me know what you do". I very quickly got my commercial web application running on it, while still accessing an MS SQL Server in my office. (For various definitions of "run" -- it was slow) Because of memory constraints, I replaced Tomcat with Jetty. But it all worked. I don't think I spent more than a couple hours on it -- after a couple of initial hours learning a bit about it, how to create a bootable SD, configure it, SSH, SCP and VNC into it to do remote deployment and configuration.
But that was a real example of moving a good sized commercial Java web application from Windows to Linux and x86/64 to ARM -- without even recompiling my binaries! (eg, java "JAR" file)
The lower I set my standards the more accomplishments I have.
(Score: 4, Interesting) by HiThere on Monday February 25 2019, @05:36PM (4 children)
Yes, but for *me* and *my application* Java is a non-starter. I wish this wasn't true, as I'd prefer choices besides Python and Ruby. All the Java serialization approaches concentrate on portability at the expense of compact data. Whoops!
Actually, C++ and D, and even go, would be reasonable choices, but Python's a lot faster to develop in. Java would be nearly as fast if it didn't discriminate so against 64 bit integers and utf8 (or even if it just handled utf32 well). Vala looks good, but it's too tied to the Gnome system. Objective-C has all their documentation specific to the Mac. So it's going to be Python, with a goal of eventually translating the code into C++ or D. (FWIW, the main problem is lists "arbitrarily" embedded into lists as the data structure. LISP would be the reasonable approach, but it seems essentially a dead language, and one can point to lots of reasons why. Clojure inherites problems from Java, and also has data all being immutable, instead of only externally visible data being immutable. Erlang would require all the mutable data to be stored in a hash table or database. Etc. Then there are the problems with available documentation systems. Fortunately DOxygen will work with Python, as Sphinx is terrible, and when they converted epydoc into PyDoc they made it so garish as to be unreadable. (It also generates truly horrible html. For one example just look at the white-space handling.)
I should also consider Fortran, but I haven't found any good text on modern Fortran, specifically including focus on concurrency. I've got a couple of bad ones. And I did consider Ada...but though it's got a few good features that don't exist in any other language I've seen, it's got so many bad features that I dropped it. (Look into writing a binary file with a header record different from the rest of the contents. It actually *is* doable, but what a pain! And to call the way of documenting your code terrible is to over-praise it dramatically.)
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
(Score: 2) by DannyB on Monday February 25 2019, @06:00PM
Wow. As you point out, there are LOTS of choices. All with pros and cons. If there were one perfect choice everyone would use it. I picked Java long ago because it seemed, and has been a very good fit building big code base web applications. Particularly refactoring tools in modern IDEs.
I still like Lisp even though, as you say, it is a dead language. Yet there is so much you learn by using it. I guess like Latin. :-) And there are a lot of fun things you can do in Lisp and related languages. Just for pure fun.
I don't know what you're trying to do, so I don't know specifically why Java is unsuitable, but it sounds like you've really understood your particular problems and workable choices.
The lower I set my standards the more accomplishments I have.
(Score: 2) by bobthecimmerian on Wednesday February 27 2019, @11:52AM (2 children)
Since I'm a Clojure fan, I'll point out that Clojure uses immutability by default but has ways you can work with mutable data. But I suspect it still doesn't fit - if Java won't work, it would be weird if an alternative built on top of Java would work. Though if I recall correctly, Clojure syntax makes handling of 64 bit uints or equivalent easier than Java. With Java the BigInteger syntax gets clunky in a hurry. Clojurescript could also be an option if you go that route, I suppose.
If you're doing something involving number-crunching, reddit programming seems to have a lot of fans of the Julia and Nim languages. I haven't used either, though.
Good luck no matter how you go.
(Score: 2) by HiThere on Wednesday February 27 2019, @05:38PM (1 child)
Julia won't work, but I do keep looking at Nim. I'm just not sure how seriously to take it.
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
(Score: 2) by HiThere on Wednesday February 27 2019, @05:57PM
So. I just looked again. Many of the documentation links are stale. (You need to recode them from the git page to the org page. Easy enough translation, but they move the site without updating the links.)
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
(Score: 4, Insightful) by jmorris on Monday February 25 2019, @05:44PM (2 children)
Yeah, but even today Java is write once, debug everywhere. Which is why anyone sane packages the exact JVM used for development and testing with the application. Which means you deploy on the same hardware arch you develop and test on unless you are a fool. Of course the IT world is largely fools now so....
The bottleneck to ARM deployment is the video problem. Small ARM boards that are more than powerful enough to do development work on exist, but they are all either entirely headless or usable only as media players because there are no stable accelerated video drivers for Linux. Android you can get EGL drivers for, but not GL and even now Vulkan is still rare.
AMD is the one hot and bothered to push ARM into the server space so they need to realize the problem and that they are uniquely suited to fixing it. Dump Mali, bolt on a medium spec Radeon core to a midgrade ARM and release a developer targeted board. Aim for a Micro-ATX form factor, put a pair of DDR4 slots for RAM, a few PCIe slots for the odd peripherals the real world requires that aren't on USB interfaces and of course some USB3.1 ports for the stuff that is. Add a GigE and call it good. Better still, once you have PCIe just put a PCIe x16 slot in and list some specific Radeon boards (a low, med and high end at minimum) that ARM support has been upstreamed for. Then price to compete with an AMD x86 of similar performance. Yea, that would mean taking a loss for a year to seed the market, AMD might have trouble doing that part.
Then embark on a PR campaign to make having one the new hotness, like they did with the PR campaign for the RaspPi.
(Score: 3, Interesting) by DannyB on Monday February 25 2019, @05:54PM
Maybe for Swing, creating desktop UIs. It doesn't seem to be the case for web applications where the Java code is mostly a "function" that takes inputs and serves up web pages. Where you see cross-platform differences that lead to bugs is when you are using things that touch platform specifics, like a desktop GUI in Swing. Or maybe even JavaFX. I have had to make minor platform adjustments for Swing, and I have not used JavaFX so I can't speak to that. Most of my Java experience is now with web server applications, and I don't see debug everywhere.
In practice, what we deploy on is pretty close to what we develop, test and demo on. (Windows, x86, the biggest variation is which version of Java or Tomcat, but the app seems highly insensitive to that variation.)
I wish AMD luck with pushing ARM into the server space. Intel needs some serious competition. My dream would be for ARM and Linux to be cost effective enough to force managers to take notice.
The lower I set my standards the more accomplishments I have.
(Score: 2) by bobthecimmerian on Wednesday February 27 2019, @11:56AM
I think the 'write once, debug everywhere is grossly overstated'. I work at a Java web shop and we have developers working on Macs, on Windows, and on Ubuntu Linux and our deployment platform is CentOS Linux and we never have a cross-platform problem. And twelve or so years ago it was "develop on Windows, deploy on Solaris" and everything was fine then too. Now if we were writing GUI applications, it might be different.
(Score: 2) by MichaelDavidCrawford on Monday February 25 2019, @06:47PM (1 child)
It's about the same as a P6 from Y2K.
I was happy with my own P6 back then; I used that box for four solid years.
Yes I Have No Bananas. [gofundme.com]
(Score: 3, Funny) by DannyB on Tuesday February 26 2019, @03:14PM
Wow. You just made me suddenly realize. A present day Rasperry Pi 3, with 1 GB of ram is:
* faster
* more memory
* much more storage (with a 32 GB or more SD card)
than my first decent Linux box in 1999.
* Athlon something or other
* 256 MB memory
* 30 GB drive
The lower I set my standards the more accomplishments I have.
(Score: 2) by zocalo on Monday February 25 2019, @12:03PM (1 child)
He's not saying ARM is dead, he's specifically saying that x86/x64 has now succesfully pwned the server market; game over. He's not ruling out that ARM might (and quite possibly will) achieve similar dominance on the embedded CPU market (including phones, telemetry devices, printers, etc.). I think he's probably right, although I do wish it were otherwise. For the right workload, I can easily imagine a Beowulf cluster of ARM boards having the lowest TCO, actually procuring and building it however is something else entirely.
UNIX? They're not even circumcised! Savages!
(Score: 2) by theluggage on Monday February 25 2019, @11:23PM
I think your comment fell through a time warp from sometime around 2000. It was right at the time. ARM has achieved dominance in the phone market, at least, and probably embedded too (given that the SSD in your PC probably contains an ARM).
Now we have a huge "cloud" industry primarily running webservers, DMBSs and scripting languages which are all already up and running on ARM Linux and pretty much processor-agnostic as far as application developers and day-to-day operations are concerned. Power consumption is a big thing and if an ARM system can deliver equivalent performance to Intel at lower power consumption, customers will listen. Intel has hit the wall on die shrink and is trying to pack the cores in instead - and that's where ARM can get them.
(Score: 2) by SomeGuy on Monday February 25 2019, @02:38PM (2 children)
I hear PowerPC is making a comeback. Those Talos II systems sound pretty sweet: https://www.raptorcs.com/TALOSII/ [raptorcs.com]
(Score: 1, Interesting) by Anonymous Coward on Monday February 25 2019, @03:21PM (1 child)
Call me cheap [hardkernel.com] but commodity parts win out.
(Score: 1, Insightful) by Anonymous Coward on Monday February 25 2019, @11:27PM
if you think backdoored motherboards and CPUs are acceptable then you are not the target demo at this time. also, they are getting more affordable. the octacore bundle is 1,575.00.
https://www.raptorcs.com/content/BK1B02/intro.html [raptorcs.com]
that's not cheap, but it's not completely terrible for a freedom respecting/security possible platform.
(Score: 3, Interesting) by urza9814 on Monday February 25 2019, @04:34PM
Yeah...I'm pretty sure I have more ARM devices in my home than x86 these days. Definitely more overall power in x86, but those are also more limited simply because all of that power is in a small number of big, multi-purpose devices. If I want to spin up a virtual Linux server, it's gonna be x86. If I want to play with an independent device, something that needs more I/O than just the network, something that I want separated from my main network for whatever reason, or if I just want it at a different physical location, it's going to be ARM.
Of course, the development that I do on ARM isn't really any different from what I would do on x86. It's mostly scripting languages -- bash, PHP, maybe a bit of Java. If I'm using C, I'm probably programming an Arduino style device.
Operating system matters more than instruction set for a lot of development work -- most of the time I really couldn't care less if I'm deploying on ARM or x86, but I'm sure as shit not deploying on a Windows server! Gimme Debian, CentOS, FreeBSD, even AIX; on ARM or x86; any of that is fine and I'll barely notice the difference. Just make sure it's *nix.