64-bit operating system

[fruitoftheloom]

It's a well worn argument, and the answer is that it'll be done when it needs to be done.

Clearly there are some performance advantages to using 64bit, and some support disadvantages. At the moment, the support disdvantages win.

So, is there any ETA when the 64 bit will be available? I understand the pros and cons analysis, but we really need the 64 bit version.

There are several ARM64 Community Supported Linux Distributions, considering the RPi 3B was released Feb 2016, the RPF / RPT have never stated they are intending to release a ARM64 Rasbian / Debian despite several Posts demanding

[jamesh]

yvette,

...we really need the 64 bit version.

I'm curious.

Could you say exactly what it is you are wanting to do with a Pi that absolutely requires a 64 bit OS?

If you really need a 64 bit OS on the Pi then there are many to choose from as suggested here already.

ISTR some databases required 64 bits for larger datasets. Not huge use case for the pi though - if you really want a database that large, there are probably better devices.

[W. H. Heydt]

So, is there any ETA when the 64 bit will be available? I understand the pros and cons analysis, but we really need the 64 bit version.

Okay...I'll bite. What is it that you "really need" a 64-bit OS for?

[W. H. Heydt]

ISTR some databases required 64 bits for larger datasets. Not huge use case for the pi though - if you really want a database that large, there are probably better devices.

My immediate reaction is that any dataset that can't be handled by a 32-bit database engine is almost certainly too big the used effectively on a Pi. Yes, you can attach multiple terabytes of storage to a Pi, but can you read through it in a reasonable length of time?

[jamesh]

ISTR some databases required 64 bits for larger datasets. Not huge use case for the pi though - if you really want a database that large, there are probably better devices.

My immediate reaction is that any dataset that can't be handled by a 32-bit database engine is almost certainly too big the used effectively on a Pi. Yes, you can attach multiple terabytes of storage to a Pi, but can you read through it in a reasonable length of time?

That was my thought. Although I am sure there will be some cases where its OK.

[Heater]

Yes. I have been toying with CockroachDB on the Pi. Not for any practical purpose except CockroadDB is a fault tolerant distributed database and firing up a cluster of Pi was the easiest, quickest thing to hand. And I was just curious as to how it would behave on a small system.

Cockroach is designed for 64 bit system. It can be built as a 32 bit program but it requires hacking around with the source and build system which I did not want to do, I wanted to build it as is.

No problem, I just used pi64.

Cockroach on the Raspbery Pi 3 - 64 bit: https://forum.cockroachlabs.com/t/cockr ... 4-bit/1246

Works fine.

Hence my question about what is it that really needs a 64 bit OS.

The only other case I know of is MongoDB. Which I have not tried.

[W. H. Heydt]

Yes. I have been toying with CockroachDB on the Pi. Not for any practical purpose except CockroadDB is a fault tolerant distributed database and firing up a cluster of Pi was the easiest, quickest thing to hand. And I was just curious as to how it would behave on a small system.

Cockroach is designed for 64 bit system. It can be built as a 32 bit program but it requires hacking around with the source and build system which I did not want to do, I wanted to build it as is.

No problem, I just used pi64.

Cockroach on the Raspbery Pi 3 - 64 bit: https://forum.cockroachlabs.com/t/cockr ... 4-bit/1246

Works fine.

Hence my question about what is it that really needs a 64 bit OS.

The only other case I know of is MongoDB. Which I have not tried.

I've read what you've said about CockroachDB with considerable interest. At present, my one active use for a database is using MySQL (though last week I set up a test/demo Pi with MariaDB...and, yes, all I had to do was import a mysqldump file to get it running, so trivially easy) but an automated distributed database would be a very nice way to go. I run the MySQL database with a replicated copy on a separate machine...but having something like that dynamically connects and mirrors the primary database with full failover capability has a lot of appeal.

And--indeed--I'm going to a convention for people who run conventions (SMOFcon...it's an SF fan insider joke) at the end of this month and should it come up I will be mentioning CockroachDB as something of interest for cons. I will, of course, also be mentioning Pis for any system where they would be appropriate.

[schmide]

You don't need anything. Logically you can build a system with 1 bit logic. An 8bit computer can page out memory and do anything albeit slowly.

The BCM2837 is a great chip. It has 64 bit capabilities which expose more registers, sandboxes, addressing modes, etc. Maybe someone wants to see how the different modes react in different situations. Technology moves forward.

I understand the need to keep things ubiquitous and simple, but the foundation is built, at some point you want to use it.

So much of this argument falls on the least common denominator. Reality is a 64 bit os is exactly that. It can run both 32 and 64bit code.

[fruitoftheloom]

You don't need anything. Logically you can build a system with 1 bit logic. An 8bit computer can page out memory and do anything albeit slowly.

The BCM2837 is a great chip. It has 64 bit capabilities which expose more registers, sandboxes, addressing modes, etc. Maybe someone wants to see how the different modes react in different situations. Technology moves forward.

I understand the need to keep things ubiquitous and simple, but the foundation is built, at some point you want to use it.

So much of this argument falls on the least common denominator. Reality is a 64 bit os is exactly that. It can run both 32 and 64bit code.

The RPF / RPT would have to support 2 Operating Systems, who pays the cost of the extra man hours required ???

How does a Charity justify this expenditure ???

How does this further the charitable aims ???

Utopian Dreams ???


https://code.overdrivenetworks.com/blog ... el-b-plus/

http://www.tal.org/tutorials/raspberry- ... bit-kernel

[schmide]

The RPF / RPT would have to support 2 Operating Systems, who pays the cost of the extra man hours required ???

How does a Charity justify this expenditure ???

How does this further the charitable aims ???

Utopian Dreams ???

How do they justify any piece of technology they put into it?

I bet 90% of the people with pies never use the GPIO pins, therefore. AX them.

Software adoption is driven by brand. RPF / RPT has a great brand and people love it. They didn't get here by not doing things.

[fruitoftheloom]

The RPF / RPT would have to support 2 Operating Systems, who pays the cost of the extra man hours required ???

How does a Charity justify this expenditure ???

How does this further the charitable aims ???

Utopian Dreams ???

How do they justify any piece of technology they put into it?

I bet 90% of the people with pies never use the GPIO pins, therefore. AX them.

Software adoption is driven by brand. RPF / RPT has a great brand and people love it. They didn't get here by not doing things.

You are not answering the questions, just attempting to justify Debian ARM64 without a cost / use analysis, heck the RPF do not even offer a Debian x86-64 for PC/MAC !

This type of post has been around for nigh on 3 years so the RPF/RPT are well aware of the pros and cons......

[schmide]

You are not answering the questions, just attempting to justify Debian ARM64 without a cost / use analysis, heck the RPF do not even offer a Debian x86-64 for PC/MAC !

This type of post has been around for nigh on 3 years so the RPF/RPT are well aware of the pros and cons......

How am I supposed to answer questions that only someone on the inside could answer? Therefore they are generally rhetorical.

Is everything a cost/use analysis? Are you asking these questions to get answers or to quiet the rhetoric?

I came here to express my want/need for something. I own 9 pies. I'm not the only one here.

[W. H. Heydt]

You don't need anything. Logically you can build a system with 1 bit logic. An 8bit computer can page out memory and do anything albeit slowly.

The BCM2837 is a great chip. It has 64 bit capabilities which expose more registers, sandboxes, addressing modes, etc. Maybe someone wants to see how the different modes react in different situations. Technology moves forward.

I understand the need to keep things ubiquitous and simple, but the foundation is built, at some point you want to use it.

So much of this argument falls on the least common denominator. Reality is a 64 bit os is exactly that. It can run both 32 and 64bit code.

I've said it before, but it probably bears repeating... I expect that, at some time in the future, Raspbian will go 64-bit. It is my personal opinion that part of the groundwork to make that feasible is for *all* "lines" within the Pi family to be 64-bit capable AND for the overwhelming majority of Pis in use to be 64-bit capable.

So what is blocking those paths now?

The Pi0/Pi0W and A-series lack 64-bit capable versions. This can be fixed fairly easily for the A-series, in that prototype Pi3A and/or Pi3A+ boards have been hinted at to exist. The Pi0/Pi0W is a far harder nut to crack as there is no obvious ARMv8 SoC to use there between the need for PoP RAM and cost constraints.

The other part of the path--Pis in use being overwhelmingly 64-bit capable--is happening to the B-series, as the Pi2Bv1.2, Pi3B and Pi3B+ are not only the boards currently being made and sold, but as the annual production rates rise, they (and any successor boards, which will surely be 64-bit capable) are now the majority of boards that have been sold. Over time, the number of remaining B-series boards in use will dwindle due to retirement, destruction (both accidental and deliberate), and replacement. Any that are left in a few years may well be running their existing OS and lack of a software upgrade path will not be a major issue. This will also probably be the path for the CM to CM3/CM3L and beyond. (And, FYI, that the CM is slated to stay in production at least until 2023 implies that 32-bit Raspbian will need to be supported for at lesat 5 years past that date, requiring that either the RPF/RPT holds off on 64-bit Raspbian until at least 2028, or supports two versions of Raspbian from some earlier date when a 64-bit version is released until the 32-bit version support can be dropped...some time after 2028.)

So really...it's all a matter of numbers. Having enough (fractionally) 64-bit capable boards out there, the need to find a path to 64-bit capable A-series and Zero-series, and the economics of only supporting a single version or finding the funds to support two versions.

I can understand the desire--not "need"--for Raspbian to be a 64-bit OS. However, I also understand why it is not economically possible to make it one at this time.

[ejolson]

Is everything a cost/use analysis?

From what I understand, the original Pi was designed to uplift a batch of computer science students by giving them a computer with which to learn basic skills before entering Cambridge. Something that fulfilled an educational need at the university level turned into a commercial success, evidently because it was innovative, interesting and generally useful for many people.

Rather than thinking about profit and sales, I would definitely encourage the focus to remain on being innovative, interesting and generally useful. In my opinion, these ideals are necessary for a successful technology company, though sometimes successful technology companies turn in to other kinds of companies. With the notable exception of IBM and Apple, most computer hardware manufacturers don't develop any kind of operating system at all. The fact that there are multiple 64-bit operating systems available for Intel-compatible PCs and ARM-based single-board computers is due to the efforts of Microsoft, Redhat, Canonical and others including Richard Stallman and Linus Torvalds.

[W. H. Heydt]

Is everything a cost/use analysis? Are you asking these questions to get answers or to quiet the rhetoric?

Yes. Everything is subject to a cost/benefit analysis. Doing anything else is an invitation to bankruptcy. Need I remind you that Eben Upton has an MBA in addition to his Comp. Sci. PhD?

I came here to express my want/need for something. I own 9 pies. I'm not the only one here.

Only 9? I have at least 9 each of some different models. (Granted, I've only got 5 Model A Pis, but those are--in Dr. Upton's words--the only Pi that was a "failure", in that only 100K were sold. Kind of ironic, actually, since the Model A was intended to be the original, basic Pi.)

Seriously... Yes. Everyone gets that you really, really want Raspbian to be a 64-bit OS. The explanation of *why* that isn't happening in the short run seem not to be getting through.

[fruitoftheloom]

Is everything a cost/use analysis? Are you asking these questions to get answers or to quiet the rhetoric?

Yes. Everything is subject to a cost/benefit analysis. Doing anything else is an invitation to bankruptcy. Need I remind you that Eben Upton has an MBA in addition to his Comp. Sci. PhD?

I came here to express my want/need for something. I own 9 pies. I'm not the only one here.

Only 9? I have at least 9 each of some different models. (Granted, I've only got 5 Model A Pis, but those are--in Dr. Upton's words--the only Pi that was a "failure", in that only 100K were sold. Kind of ironic, actually, since the Model A was intended to be the original, basic Pi.)

Seriously... Yes. Everyone gets that you really, really want Raspbian to be a 64-bit OS. The explanation of *why* that isn't happening in the short run seem not to be getting through.

Hallelujah !

[schmide]

Is everything a cost/use analysis? Are you asking these questions to get answers or to quiet the rhetoric?

Yes. Everything is subject to a cost/benefit analysis. Doing anything else is an invitation to bankruptcy. Need I remind you that Eben Upton has an MBA in addition to his Comp. Sci. PhD?

I came here to express my want/need for something. I own 9 pies. I'm not the only one here.

Only 9? I have at least 9 each of some different models. (Granted, I've only got 5 Model A Pis, but those are--in Dr. Upton's words--the only Pi that was a "failure", in that only 100K were sold. Kind of ironic, actually, since the Model A was intended to be the original, basic Pi.)

Seriously... Yes. Everyone gets that you really, really want Raspbian to be a 64-bit OS. The explanation of *why* that isn't happening in the short run seem not to be getting through.

Way to take things out of context. (practically a strawman) I was never arguing the cost/use analysis. fruitoftheloom came in and made that. My reply to that was a few questions that showed how the cost/use of GPIO made that relative. I also find it disheartening that fruitoftheloom jumps into this with a Hallelujah ! as if your reply didn't stand on its own.

I'll go further. IMO PoE ports were a huge failure. I don't fault the foundation for trying. It is the right path but as with the Model A, it is just a step in the right direction.

There are so many steps in the ecosystem that don't follow a linear path. Pinouts for example. No one expects future hats to be compatible with past boards, nor should you expect flagship software to always work on entry level boards.

BTW I get why it isn't happening. What I don't get is all the opposition to the advantages of it. If you read back in this thread there were many arguments that fell flat. Code size, addressing, etc. This is what I have talked about.

As for the boards. I was just expressing that I do invest in the ecosystem. You have more. I'm sure there are those who have even more. Maybe if we all buy more they will get enough money to maybe have 2 levels of software to go with their myriad of hardware.

[W. H. Heydt]

Is everything a cost/use analysis? Are you asking these questions to get answers or to quiet the rhetoric?

Yes. Everything is subject to a cost/benefit analysis. Doing anything else is an invitation to bankruptcy. Need I remind you that Eben Upton has an MBA in addition to his Comp. Sci. PhD?

I came here to express my want/need for something. I own 9 pies. I'm not the only one here.

Only 9? I have at least 9 each of some different models. (Granted, I've only got 5 Model A Pis, but those are--in Dr. Upton's words--the only Pi that was a "failure", in that only 100K were sold. Kind of ironic, actually, since the Model A was intended to be the original, basic Pi.)

Seriously... Yes. Everyone gets that you really, really want Raspbian to be a 64-bit OS. The explanation of *why* that isn't happening in the short run seem not to be getting through.

Way to take things out of context. (practically a strawman) I was never arguing the cost/use analysis. fruitoftheloom came in and made that. My reply to that was a few questions that showed how the cost/use of GPIO made that relative. I also find it disheartening that fruitoftheloom jumps into this with a Hallelujah ! as if your reply didn't stand on its own.

I'll go further. IMO PoE ports were a huge failure. I don't fault the foundation for trying. It is the right path but as with the Model A, it is just a step in the right direction.

There are so many steps in the ecosystem that don't follow a linear path. Pinouts for example. No one expects future hats to be compatible with past boards, nor should you expect flagship software to always work on entry level boards.

BTW I get why it isn't happening. What I don't get is all the opposition to the advantages of it. If you read back in this thread there were many arguments that fell flat. Code size, addressing, etc. This is what I have talked about.

As for the boards. I was just expressing that I do invest in the ecosystem. You have more. I'm sure there are those who have even more. Maybe if we all buy more they will get enough money to maybe have 2 levels of software to go with their myriad of hardware.

I'm going to respond a little differently because there are too many closely related points to make separate comments to each without the whole thing turning into a terminally ugly post...

I'm not exactly sure how to interpret hawaiianpi's remark. I surmise that he liked what I said and approves of it being relatively terse.

So far as I can surmise, the inclusion of GPIO pins is part of the original education mission--that of allowing the Pi to be used not just as a platform to learn to program, but to permit "physical computing" (and the associated code) to be done as well. The overall cost of installing the header is (a) a matter of deciding the cost is worth it in general, and (b) has not adversely affected the target price point. Note that the second point *does* occur on the Pi0 and Pi0W...hence the more expensive Pi0WH that does include the header pins.

Funny you should mention pinouts... The Pi has been so successful in the SBC space, that after the initial shock of its arrival on the scene causing other board makers in the same space to have to reduce prices...drastically (note that prior to the launch of the Model B Pi, the then most popular board--the Beaglebone--cost $90; the next board in that line sold for $45), but the Pi pinout has become something of an industry standard. Look at nearly any hobbyist SBC and one of the claims that is made is that the GPIO header is "Pi compatible". The Pi has *established* this pinout as the standard, so there is every reason to keep it the same going forward. I don't think the RPT/RPF wants to go down the rabbit hole that IBM did when they switched from ISA connectors to Microchannel, despite the technical superiority of the latter.

I don't think anyone is arguing that 64-bit OS is a "bad" idea. Rather, what I've read indicates that people think that any advantages are outweighed by backwards compatibility and avoiding the costs of maintaining two versions of the OS. Software may be free (as in beer), but that doesn't mean that writing and maintaining software has no cost. Beyond that, the whole point of the RPF is to funnel money into educational purposes (charity...not for profit...) rather than becoming a software house with a myriad of offerings depending on what a given processor an support.

Taking the example of CockroachDB, one could turn the question around....why don't the developers and maintainers support a 32-bit version as well as the 64-bit version. There are a lot of systems out there that are running 32-bit OSes that might want to use CockroachDB. So shouldn't there be a 32-bit version as well?

And--at the end--I'm sure there are individuals with many more boards than I have, not to mention companies that buy them by the tens of thousands. Ultimately, the point is that maintaining two versions of Raspbian would strain the RPF/RPT finances and is--at present--an avoidable expense.

Also consider that I have agreed that Raspbian will go 64-bit....some day. I just don't think that day will be soon (for the reasons I've stated). Indeed, I may not live to see it.

[ejolson]

I also find it disheartening that fruitoftheloom jumps into this with a Hallelujah !

From what I understand, hallelujah means "praise you O God." While such expressions can be used to show reverence and respect in a religious setting, the use of the same phrase for humanly purposes is sometimes considered vain. In either case, it seems a bit off topic to me and is probably best ignored.

The default operating system for Raspberry Pi is 32-bit because not all Pi's currently being sold are 64-bit compatible. Switching out the 32-bit Raspbian kernel for a 64-bit kernel can and has been done. It is interesting, but as the user land is still 32-bit there is still no way to build or execute 64-bit binaries without going to some sort of complicated multi-architecture configuration. While this is possible and has been done, you might as well rebuild all of user land to be 64-bit in the process. After making all these changes, the result is so different from current Raspbian that it might as well be given a different name, for example, Gentoo.

[Heater]

schmide,

What I don't get is all the opposition to the advantages of it

I don't see any evidence that anyone at the Pi Foundation or anywhere else is in "opposition to the advantages of it". However I sure they also see the down sides. The pros and cons have to be balanced.

If Raspbian switched to 64 bits tomorrow then what?

1) All those old Pi no longer get support and updates. Surely you can see this is a problem?

2) Unless the Pi Foundation, or somebody maintains a 32 bit version as well. That takes time, effort and/or expense. Surely you can see that can be an issue.

I feel this will resolve itself when either a way to support both 32 and 64 bit builds is found that is practical.

Or when it is deemed that old Pi are old enough to stop supporting them.

In the mean time, there are a number of 64 bit Linux for the Pi around so I don't see what all the whining is about. Do like I do at just get on with it!

[Heater]

W. H. Heydt,

Taking the example of CockroachDB, one could turn the question around....why don't the developers and maintainers support a 32-bit version as well as the 64-bit version. There are a lot of systems out there that are running 32-bit OSes that might want to use CockroachDB. So shouldn't there be a 32-bit version as well?

I somehow feel you are not serious about that question. But for the benefit of others I would say that CockroachDB is intended to be a database for freakin' big distributed databases. The world of gigantic racks of servers each with dozens of gigs of RAM and dozens of processor cores. A PC or laptop with 4GB of RAM is an entry level system today and enough to get Cockroach running for some local development. That fact that it runs at all on the diminutive Pi is a bonus. There is no motivation for the Cockroach guys to worry about old Pi's anymore than getting Cockroach to run on a Z80.

You remind me, I should get back to my Cockroach on Pi cluster experiments. They kind of got interrupted by work and life since last Christmas. One thing though:

...having something like that dynamically connects and mirrors the primary database with full failover capability has a lot of appeal.

The way Cockroach is architected there is no concept of "primary" database and fail over as such. Rather all nodes act as peers. Updates can be made to any connection to any peer and they are replicated from there. There is a consensus algorithm going on to ensure everything is consistent all the time (The Raft algorithm). This requires a majority of nodes to be up and running and agree on the state of the DB. In a 3 node cluster, for example, two machines have to be working. If there is only one machine up (or it is disconnected from the other two) no reads or writes can be made there, it has no way to achieve a consensus on the consistency of the distributed data store.

I think you should have a look at Cockroach. It's pretty easy to get started with. The documentation is pretty good. Their forum as been very helpful. I found it a lot of fun killing off nodes at random and watching things continue working with out a hitch.

[rpdom]

If Raspbian switched to 64 bits tomorrow then what?

1) All those old Pi no longer get support and updates. Surely you can see this is a problem?

Seeing as those "old" Pi include the current Zero range, I suspect they will be supported for a few more years.

Whether this means that there will be two versions of Raspbian or not, well, we'll just have to wait and see.

[W. H. Heydt]

If Raspbian switched to 64 bits tomorrow then what?

1) All those old Pi no longer get support and updates. Surely you can see this is a problem?

Seeing as those "old" Pi include the current Zero range, I suspect they will be supported for a few more years.

Whether this means that there will be two versions of Raspbian or not, well, we'll just have to wait and see.

It also includes the--currently available--A+ and B+, too. And personally, I have a whole bunch of Pi2Bv1.1 boards in use.

[Gavinmc42]

One day perhaps Zero's will use dual 64bit A35 cores with VC5?
Trying to think of a use for 64bit databases on Zeros?
If it has to be small then a CM3?

For the main purpose Pi's are used for, 32bit is ok for kids.
Universities and the rest of us 64bit?
We are not going to get much performance improvement as the SDRAM is only 32bit wide.

That said, Gentoo64 is a delight to learn 64bit coding on.
Lots of 64bit learning to still to be done, Mesa3D, NEON, ML/NN/CV, Arm Compute etc.
Operative word for me is "learning", I'm not 24/7 using 64bit yet.

Now what is CockroachDB, and what can I do with it?
Hmm, fault tolerant Cloud databases? on Pi's?
Whoops, Heater already posted on here, must be getting old.
Does it run on Gentoo64?

[code_exec]

One day perhaps Zero's will use dual 64bit A35 cores with VC5?
Trying to think of a use for 64bit databases on Zeros?
If it has to be small then a CM3?

For the main purpose Pi's are used for, 32bit is ok for kids.
Universities and the rest of us 64bit?
We are not going to get much performance improvement as the SDRAM is only 32bit wide.

That said, Gentoo64 is a delight to learn 64bit coding on.
Lots of 64bit learning to still to be done, Mesa3D, NEON, ML/NN/CV, Arm Compute etc.
Operative word for me is "learning", I'm not 24/7 using 64bit yet.

Now what is CockroachDB, and what can I do with it?
Hmm, fault tolerant Cloud databases? on Pi's?
Whoops, Heater already posted on here, must be getting old.
Does it run on Gentoo64?

64-bit OSes are targeted at users with at least 3GB RAM. Since the Pi 3 only has 1GB, 64-bit OSes can run, but not very well performance wise.

However, 64-bit on the Pi is possible. These links may help you:

https://wiki.ubuntu.com/ARM/RaspberryPi ... ISO_images
https://github.com/bamarni/pi64

As for Gentoo64, according to the Gentoo Wiki, the RPi3 can indeed run Gentoo64.

https://wiki.gentoo.org/wiki/Raspberry_ ... it_Install