Re: ppc64el porter situation
On Sat, Oct 22, 2016 at 09:01:26PM +0300, Adrian Bunk wrote: > Are they developing new powerpc products? AMCC? I have no idea. Maybe not. Freescale certainly seems to be. > Their latest products are also pretty ARM. That they are. > Are you talking about new e6500 SoCs, or are you only talking about > support for existing products? New ones. > I have no doubt they will continue to provide support for e6500 for > several years, just like they supported SoCs with SPE cpus in their > SDK until December 2015. > > They released only two e6500 based SoCs for QorIQ (T2080 and T4240),[1] > and for one of them samples of ARM replacements are already available. They are not intended as replacements, just alternatives. Some people have powerpc code and see no reason to port to a little endian platform (which can be a pain in the rear). Some would rather have more options for CPUs and would like ARM. The IO bits seem to be identical between the powerpc and arm chips, so at least that only needs to be developed once. Now I suppose if demand disappears they will change their plans, but at least the place I was working until a few months ago, we were told that they were going to continue designing new powerpc and arm chips. We explicitly asked if powerpc was a dead end going forward (we were already using freescale ppc chips in some designs, and arm chips in other designs, so either was an option for us). > These are anyway big endian, but the general situation is that there > is not much powerpc development left that does not depend on IBM. I do agree that people are not running little endian on the freescale chips. -- Len Sorensen
Re: ppc64el porter situation
On Fri, Oct 21, 2016 at 01:30:13PM -0400, Lennart Sorensen wrote: > On Thu, Oct 20, 2016 at 11:18:39PM +0300, Adrian Bunk wrote: > > Freescala/NXP is not even on the OpenPOWER member list - this is not the > > old power.org > > Neither is AMCC as far as I can tell. Doesn't mean they aren't still > doing powerpc. Are they developing new powerpc products? Their latest products are also pretty ARM. > > For their network processors Freescala/NXP is moving away from PowerPC, > > and their first ARM based network processors are already on the market. > > That's not what they are telling their customers. They insist they are > very much behind both arm and powerpc. Are you talking about new e6500 SoCs, or are you only talking about support for existing products? I have no doubt they will continue to provide support for e6500 for several years, just like they supported SoCs with SPE cpus in their SDK until December 2015. They released only two e6500 based SoCs for QorIQ (T2080 and T4240),[1] and for one of them samples of ARM replacements are already available. These are anyway big endian, but the general situation is that there is not much powerpc development left that does not depend on IBM. > Len Sorensen cu Adrian [1] with T2081/T4160/T4080 variants, plus two in the Qonverge platform -- "Is there not promise of rain?" Ling Tan asked suddenly out of the darkness. There had been need of rain for many days. "Only a promise," Lao Er said. Pearl S. Buck - Dragon Seed
Re: ppc64el porter situation
On Thu, Oct 20, 2016 at 11:18:39PM +0300, Adrian Bunk wrote: > Freescala/NXP is not even on the OpenPOWER member list - this is not the > old power.org Neither is AMCC as far as I can tell. Doesn't mean they aren't still doing powerpc. > For their network processors Freescala/NXP is moving away from PowerPC, > and their first ARM based network processors are already on the market. That's not what they are telling their customers. They insist they are very much behind both arm and powerpc. -- Len Sorensen
Re: ppc64el porter situation
On Thu, Oct 20, 2016 at 03:58:21PM -0400, Lennart Sorensen wrote: > On Thu, Oct 20, 2016 at 03:54:43PM -0400, Lennart Sorensen wrote: > > I think Freescala/NXP might disagree. Not sure if the e6500 core could > > ruin ppc64el or not, but they certainly make a lot of powerpc chips. > > That should have said 'run' not 'ruin'. That would have been rather > interesting otherwise. Freescala/NXP is not even on the OpenPOWER member list - this is not the old power.org For their network processors Freescala/NXP is moving away from PowerPC, and their first ARM based network processors are already on the market. > Len Sorensen cu Adrian -- "Is there not promise of rain?" Ling Tan asked suddenly out of the darkness. There had been need of rain for many days. "Only a promise," Lao Er said. Pearl S. Buck - Dragon Seed
Re: ppc64el porter situation
On Thu, Oct 20, 2016 at 03:54:43PM -0400, Lennart Sorensen wrote: > I think Freescala/NXP might disagree. Not sure if the e6500 core could > ruin ppc64el or not, but they certainly make a lot of powerpc chips. That should have said 'run' not 'ruin'. That would have been rather interesting otherwise. -- Len Sorensen
Re: ppc64el porter situation
On Thu, Oct 20, 2016 at 10:22:14PM +0300, Adrian Bunk wrote: > https://en.wikipedia.org/wiki/MeeGo#Companies_supporting_the_project > > That's also an impressive list of companies, isn't it? > When the one company that mattered switched to a different platform, > the whole platform collapsed immediately. > > The whole Power platform also seems to be mostly around IBM. I think Freescala/NXP might disagree. Not sure if the e6500 core could ruin ppc64el or not, but they certainly make a lot of powerpc chips. -- Len Sorensen
Re: ppc64el porter situation
On Wed, Oct 19, 2016 at 12:06:59PM +0200, Aurelien Jarno wrote: > Hi, Hi Aurelien, >... > To me it looks like they are really skilled for that job. Do you have > actual facts showing the contrary? Niels said that I shouldn't hesitate to let the release team know when I believe there is an issue they have overlooked.[1] I do believe that there is a risk in the ppc64el port in the unlikely case that IBM suddenly moves away from PowerPC, and that is currently not mentioned in the architecture requalification table. This is not about past work done by the ppc64el porters, this is about a specific single point of failure that could happen in the future. In my opinion this is a risk, and the release team should be aware of it when making the decision regarding architectures in stretch. I do appreciate the answers from Breno and you that addressed some of the things I brought up. I do expect the release team to read this discussion and take it into consideration. I am not involved in the decision regarding ports in stretch, and any decision is fine with me. > Aurelien cu Adrian [1] https://lists.debian.org/debian-release/2016/10/msg00131.html -- "Is there not promise of rain?" Ling Tan asked suddenly out of the darkness. There had been need of rain for many days. "Only a promise," Lao Er said. Pearl S. Buck - Dragon Seed
Re: ppc64el porter situation
On Wed, Oct 19, 2016 at 09:06:14AM -0200, Breno Leitao wrote: > Hello Adrian, Hi Breno, > Let me share my view as the only DD listed as ppc64el porter. thanks for your reply. Just to state it explicitely in case that was not clear, I do not have any problem with you personally or the ppc64el port in general. I am just saying that I see a risk for the ppc64el port in the unlikely case that IBM makes a sudden move away from PowerPC during the lifetime of stretch. > On Mon, Oct 17, 2016 at 10:50:01PM +0300, Adrian Bunk wrote: > > Is a DM enough, if the only DD gets killed by a car [2] the day after > > the release of stretch? > > The other DM is in the process of becoming a DD[1]. This might reduce > the truck factor by half. > > [1] https://nm.debian.org/person/frediz That's good news. > > Second, all 4 committed porters seem to be employees of IBM. > > > > What happens if for whatever good or bad reason IBM decides in 2018 > > or 2019 to go away from ppc64el, and all 4 committed porters get fired? > > I understand your point here. ppc64el architecture is IBM's current and > future focus. ppc64el is also planned for POWER9 and beyond. While it's > hard to predict what future business decisions IBM may make, we believe > the future of ppc64el and OpenPower systems looks good. > > There are many other distros that support ppc64el at this moment, as > Ubuntu, Fedora, SLES, RHEL and others coming. So, your point is not > Debian specific, but, generic to the Linux ecossystem. Debian is in a different situation, the porters of these distributions are likely employed by the company behind the distribution and not by IBM. > > The wording of the porter commitment is already limited to "I intend > > to", and there is the single point of failure that one business > > decision by IBM might reduce the number of porters immediately from 4 > > to 0. > > Right, since ppc64el machines are not desktop/personal machines, it is > harder to get porters, compared to more pervasive architectures, as amd64. > I hope to have more DD porters in the future, as ppc64el become more > prevalent. > > lso, there are many other hardware manufactors and partners that relies > on Linux for the Power platform[1]. In my opinion, the Power platform is > bigger than IBM at this moment. > > [1] http://openpowerfoundation.org/membership/current-members/ https://en.wikipedia.org/wiki/MeeGo#Companies_supporting_the_project That's also an impressive list of companies, isn't it? When the one company that mattered switched to a different platform, the whole platform collapsed immediately. The whole Power platform also seems to be mostly around IBM. >... > On the other side, if there is a requirements for being a porter that > says that the porter might be able to fix difficult issues on kernel and > toolchain, then it is a different story. I do not believe that this > requirement exists. > It is not a requirement for every porter, but that skill is required. Debian got burned in wheezy in the sparc port when no porter was available to fix a broken kernel after the release. That was an embarrassment to the Debian stability and quality that noone wants to ever see again. cu Adrian -- "Is there not promise of rain?" Ling Tan asked suddenly out of the darkness. There had been need of rain for many days. "Only a promise," Lao Er said. Pearl S. Buck - Dragon Seed
Re: ppc64el porter situation
Hello Adrian, Let me share my view as the only DD listed as ppc64el porter. On Mon, Oct 17, 2016 at 10:50:01PM +0300, Adrian Bunk wrote: > Is a DM enough, if the only DD gets killed by a car [2] the day after > the release of stretch? The other DM is in the process of becoming a DD[1]. This might reduce the truck factor by half. [1] https://nm.debian.org/person/frediz > Second, all 4 committed porters seem to be employees of IBM. > > What happens if for whatever good or bad reason IBM decides in 2018 > or 2019 to go away from ppc64el, and all 4 committed porters get fired? I understand your point here. ppc64el architecture is IBM's current and future focus. ppc64el is also planned for POWER9 and beyond. While it's hard to predict what future business decisions IBM may make, we believe the future of ppc64el and OpenPower systems looks good. There are many other distros that support ppc64el at this moment, as Ubuntu, Fedora, SLES, RHEL and others coming. So, your point is not Debian specific, but, generic to the Linux ecossystem. > The wording of the porter commitment is already limited to "I intend > to", and there is the single point of failure that one business > decision by IBM might reduce the number of porters immediately from 4 > to 0. Right, since ppc64el machines are not desktop/personal machines, it is harder to get porters, compared to more pervasive architectures, as amd64. I hope to have more DD porters in the future, as ppc64el become more prevalent. lso, there are many other hardware manufactors and partners that relies on Linux for the Power platform[1]. In my opinion, the Power platform is bigger than IBM at this moment. [1] http://openpowerfoundation.org/membership/current-members/ > Third, the skills of the committed porters for post-release work. > > It is extremely valuable when people are doing manual and automated > testing and fix the usual porting issues prior to the release. > > But the most important skills required post-release until end-2020 are > quite different. > > How many of the committed ppc64el porters are personally able to fix > difficult issues that require intimate knowledge of hardware, kernel > and toolchain? I understand that it is going to be hard to find a developer that is able to fix difficult issues on kernel/toolchain, we are relying and supported by differents team doing Kernel, virtualization, toolchain, optimization, etc. On situations we are not skilled enough to work on, we have these other teams support. This is shown by the amount of package that was ported to ppc64el[2]. We, the Debian porters, didn't do it by ourself, but we counted on different teams doing their work on each area, as from package optimizations to toolchain enablement. On the other side, if there is a requirements for being a porter that says that the porter might be able to fix difficult issues on kernel and toolchain, then it is a different story. I do not believe that this requirement exists. [2] https://buildd.debian.org/stats/ Also I assure you I have personal interest in Debian success
Re: ppc64el porter situation
Hi, On 2016-10-17 22:50, Adrian Bunk wrote: > Disclaimer: > I am not a member of the release team, and I am only speaking for myself. > > > The architecture requalification status for stretch [1] lists the > ppc64el porter situation as green, but there are three reasons why > the situation doesn't look that good to me. > > > First, official status of the porters: > - 1 DD > - 1 DM > - 2 no DD/DM > > Is a DM enough, if the only DD gets killed by a car [2] the day after > the release of stretch? I am actually not sure it makes a lot of difference being DD or DM or even not a DD/DM. What is important is that issues are fixed, that patches are provided. For that you need access to the knowledge and access to the hardware, not upload or vote rights. > Second, all 4 committed porters seem to be employees of IBM. That's true, but they are from two different teams in two different countries. > What happens if for whatever good or bad reason IBM decides in 2018 > or 2019 to go away from ppc64el, and all 4 committed porters get fired? > > The wording of the porter commitment is already limited to "I intend to", > and there is the single point of failure that one business decision > by IBM might reduce the number of porters immediately from 4 to 0. I think it's unlikely to happen, and even if that happens the porters might continue to work on Debian on their personal time. > Third, the skills of the committed porters for post-release work. > > It is extremely valuable when people are doing manual and automated > testing and fix the usual porting issues prior to the release. > > But the most important skills required post-release until end-2020 are > quite different. > > How many of the committed ppc64el porters are personally able to fix > difficult issues that require intimate knowledge of hardware, kernel > and toolchain? The 4 porters have been working on ppc64el for years, they have done the initial bootstrapping outside of Debian, they have done the initial bootstrap in Debian, they have participated in the release of Jessie and they have sent hundred of patches in the BTS. To me it looks like they are really skilled for that job. Do you have actual facts showing the contrary? Aurelien -- Aurelien Jarno GPG: 4096R/1DDD8C9B aurel...@aurel32.net http://www.aurel32.net signature.asc Description: PGP signature
