So Power is switching to little endian by default?
Only slightly related, but I’ve been looking for somewhere to test a piece of code on big endian, but that seems to be rather difficult as a private person. I think the only options are to find some physical hardware on the cheap?
I have a Pi3, and that’s supposed to be bi-endian, but I’m not sure how to go about installing a big endian Linux on it. Same goes for a Scaleway ARM virtual machine, I guess.
Shell accounts at Polarhome are free for developers of open source projects (and cheap otherwise). Their Debian/PPC and Solaris/SPARC are big-endian IIRC.
You should be able to virtualize, Debian for example supports some Big Endian architectures. I don’t reckon it matters much though, Big Endian is definitely on the way out.
If you do want to go physical, you can get an Octeon-based system, they’re Big Endian mips64. Mostly used in networking equipment. Cavium has an incomplete list of products using Octeon processors, stuff under the consumer tab is probably your best bet for cheap stuff.
I have a Ubiquiti UniFi Security Gateway running on Octeon. It’s running some kind of Debian derivative, or so I assume since dpkg and the Debian package keys are present.
$ lscpu
Architecture: mips64
Byte Order: Big Endian
[...]
$ uname -a
Linux ubnt 3.10.20-UBNT #1 SMP Fri Nov 3 15:45:37 MDT 2017 mips64 GNU/Linux
Yeah, Ubiquiti’s Octeon stuff (specifically EdgeRouter) is quite well known, it’s supported by FreeBSD and OpenBSD for example. But consumer router grade CPUs are uhhhh rather weak :(
It’s still very prohibitively expensive unless you’re very dedicated to having a POWER box. I have access to off-lease POWER6 boxes acquired for cheap on eBay, but those are large, loud, pour out heat, suck up electricity, and generally only desirable if you really want a POWER box but lack funds. (Not to mention the firmware bugs that IBM refused to patch for it, so newer distros don’t support POWER6.)
Really, the best way to play with PPC still is to buy an old Power Mac, which is kinda sad.
But I’m going to get it because it’s all FOSS, no blobs, that’s the main reason. It’s also not that expensive, considering specs. And it’s just as power hungry as similar Intel boxes. Sure, older POWER generations were much more power hungry, but things changed with POWER9.
I got quality time with Power7 and Power8 infrastructure when I was at IBM, and I can’t say enough good things about performance, stability, etc. But I don’t think they’ve got the economics right. There really needs to be a true entry level machine, something on par with an HP DL360 or Dell R400 type, something in the sub-$2k price range, that would induce a substantial number of users say “I can afford to pick one or two up just to check it out”. They’re simply way outside the kick-the-tires price point to make adoption easy.
A small and cheaper machine might be good for i users as well - a large chunk of them of them are still hanging onto their old AS/400 Model 150/170s running a hopelessly old version of the OS, and “gone off the grid.” IBM shows little interest in trying to get them back on the wagon again; “entry level” POWER9 is still very, very expensive and very large compared to a tower-sized Model 150. A small POWER9 wouldn’t get just new customers, but it’d turn long-time shops back into paying customers again.
(It has to be an IBM design too; they’re the only ones with tagged memory extensions that i needs.)
I might be missing something here, but to my knowledge there’s never been an inexpensive i-series version of Power. So even if they release a sub-us$2k Power9 Linux/AIX box, the AS/400 division is never going to chase that market, even if that might be a smart move.
Probably because they were an all-in-one solution marketed to replace a pile of servers, enterprise DB integrated last I checked, and not needing much administration. That’s the kind of thing one can avoid making a commodity for a while. We know it would be a smart move to chase that market with cheap offerings due to Net Integrators’ Nitix boxes that did similar things with UNIX tech. Really neat development back then. What happened to them? IBM acquired them, shelved the regular solution, and re-released it as something in Lotus portfolio. Typical IBM… (sighs)
You keep using past tense there. AS/400 nee i-Series is now almost 30 years old (40 years old if you consider the System/3x ancestors) and is still a profitable line of business for IBM. Apparently, they’ve managed to not make it a commodity for a long time. They’ve got an awfully long track record of success for us to second guess them and say they need a dirt cheap/hobbyist/tire kicker AS/400; that smacks of hubris.
The AS/400 doesnt have same feature set as an IBM i. The System/38 similarly had differences to AS/400. I said were because AS/400 is an older product in ths family. Far as will never, people said that about mainframes, too, long ago. IBM has entry-level ones now that let people experiment for a fraction of a real mainframe.
I dont think they want to do IBM i that way but might in future.
“They’ve got an awfully long track record of success for us to second guess them”
You could use same logic to say COBOL is still a huge, profitable language because it’s superior to modern ones. Instead, IBM tech and legacy systems have a high switching cost that keeps customers upgrading their boxes instead of porting to Linux or something. Vendor lockin. IBM also cleverly let the newer ones run Linux in VM’s to reduce some motivation for porting. That the machines are also incredibly reliable compared to risks of a switch look even higher.
If there was no cost for a port, you bet a lot of businesses would consider moving their AS/400 apps onto a highly-available, mostly-self-managing set of x86 machines running enterprise Linux. Esp when they look at $2-5k that Nitix-like solutions did/could cost vs whatever IBM i’s go for now. That many were already doing new developments on Linux or attempting ports via migration companies likely contributed to IBM supporting it in IBM i systems.
The i-Series is upward compatible with AS/400, and if it “doesn’t have the same feature set”, it doesn’t in the same way that Solaris 2.1 had a different feature set that Solaris 2.8. Yes, in 30 years, new features hopefully creep in. And yes, IBM has entry-level z-Series…but they aren’t sub-$2k items. Other than that, I don’t really see what you said that contradicts what I did.
Looking at @calvin’s link, I found two resources in the comments for people trying to ensure FOSS works on POWER boxes or mainframes. They basically share boxes and advice with developers.
Of course, without IBM selling affordable boxes, anyone doing this outside of a business selling to POWER/mainframe customers will be jumping through hoops to get software working on platforms bringing in 8+ zeros for IBM. As in, they’re benefiting heavily off of this work but can’t seem to reward FOSS developers with access to cheap or at-cost hardware in return. I’m just mentioning this in event some people have ethical objections to putting in effort for ISA of a company going out of their way to limit access/adoption. The approach I took was just buying a used Mac laptop off eBay. They’re $70 for a portable, 1GHz PPC4 setup. They eat through the battery quickly, too.
One more option not mentioned are embedded boards. While IBM moved on with POWER, the PPC segment stayed in embedded with all kinds of CPU’s, hardware accelerators, peripherals, etc. Freescale is leading supplier of these. The safety-critical industry that runs things like separation kernels heavily uses PPC boards. Some are quite expensive but there could be cheap ones, too. Especially used. Similarly, Xilinx puts PPC cores into some of its Virtex FPGA’s. Might better justify buying PPC core if you were already going to spend per-unit pricing to get a FPGA. Some examples to illustrate features.
Note: PowerQUICC III’s were the ones I was eyeballing for PPC desktop before Raptor was proposed. They would still be easier to build since product is already available but obviously we’d all love POWER9’s if that product/price can happen.
So Power is switching to little endian by default?
Only slightly related, but I’ve been looking for somewhere to test a piece of code on big endian, but that seems to be rather difficult as a private person. I think the only options are to find some physical hardware on the cheap?
I have a Pi3, and that’s supposed to be bi-endian, but I’m not sure how to go about installing a big endian Linux on it. Same goes for a Scaleway ARM virtual machine, I guess.
Shell accounts at Polarhome are free for developers of open source projects (and cheap otherwise). Their Debian/PPC and Solaris/SPARC are big-endian IIRC.
You can also run QEMU, here’s a random repo with instructions.
You should be able to virtualize, Debian for example supports some Big Endian architectures. I don’t reckon it matters much though, Big Endian is definitely on the way out.
If you do want to go physical, you can get an Octeon-based system, they’re Big Endian mips64. Mostly used in networking equipment. Cavium has an incomplete list of products using Octeon processors, stuff under the consumer tab is probably your best bet for cheap stuff.
I have a Ubiquiti UniFi Security Gateway running on Octeon. It’s running some kind of Debian derivative, or so I assume since
dpkgand the Debian package keys are present.This seems consistent with the development kit information on the Cavium Octeon web page:
My other UniFi hardware runs Little Endian ARMv7 though. Looks like processors made by either MediaTek, or Qualcomm for the wireless gizmos.
Yeah, Ubiquiti’s Octeon stuff (specifically EdgeRouter) is quite well known, it’s supported by FreeBSD and OpenBSD for example. But consumer router grade CPUs are uhhhh rather weak :(
Or just get actual POWER box. Talos II (mentioned in the article) is relatively cheap for the specs.
It’s still very prohibitively expensive unless you’re very dedicated to having a POWER box. I have access to off-lease POWER6 boxes acquired for cheap on eBay, but those are large, loud, pour out heat, suck up electricity, and generally only desirable if you really want a POWER box but lack funds. (Not to mention the firmware bugs that IBM refused to patch for it, so newer distros don’t support POWER6.)
Really, the best way to play with PPC still is to buy an old Power Mac, which is kinda sad.
edit: interesting thread on this topic of high-end RISC systems being hard to acquire for devs, which reduces their viability on the market
I guess I am dedicated :D
But I’m going to get it because it’s all FOSS, no blobs, that’s the main reason. It’s also not that expensive, considering specs. And it’s just as power hungry as similar Intel boxes. Sure, older POWER generations were much more power hungry, but things changed with POWER9.
the IBM PDP program gives access to POWER based systems, they’ve just added 9 support but previously had 7 & 8 based systems running AIX & Suse.
I got quality time with Power7 and Power8 infrastructure when I was at IBM, and I can’t say enough good things about performance, stability, etc. But I don’t think they’ve got the economics right. There really needs to be a true entry level machine, something on par with an HP DL360 or Dell R400 type, something in the sub-$2k price range, that would induce a substantial number of users say “I can afford to pick one or two up just to check it out”. They’re simply way outside the kick-the-tires price point to make adoption easy.
A small and cheaper machine might be good for i users as well - a large chunk of them of them are still hanging onto their old AS/400 Model 150/170s running a hopelessly old version of the OS, and “gone off the grid.” IBM shows little interest in trying to get them back on the wagon again; “entry level” POWER9 is still very, very expensive and very large compared to a tower-sized Model 150. A small POWER9 wouldn’t get just new customers, but it’d turn long-time shops back into paying customers again.
(It has to be an IBM design too; they’re the only ones with tagged memory extensions that i needs.)
I might be missing something here, but to my knowledge there’s never been an inexpensive i-series version of Power. So even if they release a sub-us$2k Power9 Linux/AIX box, the AS/400 division is never going to chase that market, even if that might be a smart move.
Probably because they were an all-in-one solution marketed to replace a pile of servers, enterprise DB integrated last I checked, and not needing much administration. That’s the kind of thing one can avoid making a commodity for a while. We know it would be a smart move to chase that market with cheap offerings due to Net Integrators’ Nitix boxes that did similar things with UNIX tech. Really neat development back then. What happened to them? IBM acquired them, shelved the regular solution, and re-released it as something in Lotus portfolio. Typical IBM… (sighs)
You keep using past tense there. AS/400 nee i-Series is now almost 30 years old (40 years old if you consider the System/3x ancestors) and is still a profitable line of business for IBM. Apparently, they’ve managed to not make it a commodity for a long time. They’ve got an awfully long track record of success for us to second guess them and say they need a dirt cheap/hobbyist/tire kicker AS/400; that smacks of hubris.
The AS/400 doesnt have same feature set as an IBM i. The System/38 similarly had differences to AS/400. I said were because AS/400 is an older product in ths family. Far as will never, people said that about mainframes, too, long ago. IBM has entry-level ones now that let people experiment for a fraction of a real mainframe.
I dont think they want to do IBM i that way but might in future.
“They’ve got an awfully long track record of success for us to second guess them”
You could use same logic to say COBOL is still a huge, profitable language because it’s superior to modern ones. Instead, IBM tech and legacy systems have a high switching cost that keeps customers upgrading their boxes instead of porting to Linux or something. Vendor lockin. IBM also cleverly let the newer ones run Linux in VM’s to reduce some motivation for porting. That the machines are also incredibly reliable compared to risks of a switch look even higher.
If there was no cost for a port, you bet a lot of businesses would consider moving their AS/400 apps onto a highly-available, mostly-self-managing set of x86 machines running enterprise Linux. Esp when they look at $2-5k that Nitix-like solutions did/could cost vs whatever IBM i’s go for now. That many were already doing new developments on Linux or attempting ports via migration companies likely contributed to IBM supporting it in IBM i systems.
The i-Series is upward compatible with AS/400, and if it “doesn’t have the same feature set”, it doesn’t in the same way that Solaris 2.1 had a different feature set that Solaris 2.8. Yes, in 30 years, new features hopefully creep in. And yes, IBM has entry-level z-Series…but they aren’t sub-$2k items. Other than that, I don’t really see what you said that contradicts what I did.
Looking at @calvin’s link, I found two resources in the comments for people trying to ensure FOSS works on POWER boxes or mainframes. They basically share boxes and advice with developers.
OSU’s POWER Lab
http://osuosl.org/services/powerdev/
Open Mainframe Project
https://www.openmainframeproject.org/blog/2018/01/08/make-mainframe-support-new-years-resolution-open-source-project-2018
Of course, without IBM selling affordable boxes, anyone doing this outside of a business selling to POWER/mainframe customers will be jumping through hoops to get software working on platforms bringing in 8+ zeros for IBM. As in, they’re benefiting heavily off of this work but can’t seem to reward FOSS developers with access to cheap or at-cost hardware in return. I’m just mentioning this in event some people have ethical objections to putting in effort for ISA of a company going out of their way to limit access/adoption. The approach I took was just buying a used Mac laptop off eBay. They’re $70 for a portable, 1GHz PPC4 setup. They eat through the battery quickly, too.
One more option not mentioned are embedded boards. While IBM moved on with POWER, the PPC segment stayed in embedded with all kinds of CPU’s, hardware accelerators, peripherals, etc. Freescale is leading supplier of these. The safety-critical industry that runs things like separation kernels heavily uses PPC boards. Some are quite expensive but there could be cheap ones, too. Especially used. Similarly, Xilinx puts PPC cores into some of its Virtex FPGA’s. Might better justify buying PPC core if you were already going to spend per-unit pricing to get a FPGA. Some examples to illustrate features.
https://www.nxp.com/products/processors-and-microcontrollers/power-architecture-processors:POWER-ARCHITECTURE
https://en.wikipedia.org/wiki/List_of_PowerPC_processors#Embedded_PowerPC
http://phytec.com/products/system-on-modules/phycore/mpc5200b-io/#som
https://www.ebay.com/i/181791876772?chn=ps&dispItem=1
Note: PowerQUICC III’s were the ones I was eyeballing for PPC desktop before Raptor was proposed. They would still be easier to build since product is already available but obviously we’d all love POWER9’s if that product/price can happen.