Hello everyone,
I am thinking about building custom kernels for CentOS-7 with a support for Multipath-TCP (http://multipath-tcp.org/), and I am wondering if a such heavy modification should have its place in the centosplus kernel?
Modified kernel sources are available at https://github.com/multipath-tcp/mptcp and equivalent patches are there: http://multipath-tcp.org/patches/.
Best regards, Guillaume Derval.
On 06/15/2014 12:54 PM, Guillaume Derval wrote:
Hello everyone,
I am thinking about building custom kernels for CentOS-7 with a support for Multipath-TCP (http://multipath-tcp.org/), and I am wondering if a such heavy modification should have its place in the centosplus kernel?
sure, that is the sort of thing that works in the plus kernel best. Akemi ?
On Sun, Jun 15, 2014 at 4:17 PM, Karanbir Singh mail-lists@karan.org wrote:
On 06/15/2014 12:54 PM, Guillaume Derval wrote:
Hello everyone,
I am thinking about building custom kernels for CentOS-7 with a support for Multipath-TCP (http://multipath-tcp.org/), and I am wondering if a such heavy modification should have its place in the centosplus kernel?
sure, that is the sort of thing that works in the plus kernel best. Akemi ?
Adding patches is absolutely one of the things that the plus kernel offers. In this case, as Guillaume stated, the modifications are quite extensive. Without understanding all the details of the patches, I would like to ask if there are any consequences that might "negatively" affect the existing plus kernel users.
Akemi
On 06/16/2014 01:31 AM, Akemi Yagi wrote:
On Sun, Jun 15, 2014 at 4:17 PM, Karanbir Singh mail-lists@karan.org wrote:
On 06/15/2014 12:54 PM, Guillaume Derval wrote:
Hello everyone,
I am thinking about building custom kernels for CentOS-7 with a support for Multipath-TCP (http://multipath-tcp.org/), and I am wondering if a such heavy modification should have its place in the centosplus kernel?
sure, that is the sort of thing that works in the plus kernel best. Akemi ?
Adding patches is absolutely one of the things that the plus kernel offers. In this case, as Guillaume stated, the modifications are quite extensive. Without understanding all the details of the patches, I would like to ask if there are any consequences that might "negatively" affect the existing plus kernel users.
its certainly a role that is very interesting - and maybe a middle-initial-ground might be to carry a multipath-tcp specific kernel, just to bootstrap things off ?
Hello everyone,
On 06/16/2014 01:31 AM, Akemi Yagi wrote:
Adding patches is absolutely one of the things that the plus kernel offers. In this case, as Guillaume stated, the modifications are quite extensive. Without understanding all the details of the patches, I would like to ask if there are any consequences that might "negatively" affect the existing plus kernel users.
As I am not a developer of Multipath-TCP, I asked the main developer (Christoph Paasch) about that. Here is what he answered (translated from french):
I see no « negative consequences » in the inclusion of MPTCP in the centosplus kernel. Everything should work normally (and if its not, it’s a bug we can fix). The impact on the performance of plain TCP should be very low.
Best regards, Guillaume Derval.
On Mon, Jun 16, 2014 at 2:41 AM, Guillaume Derval guillaume.derval@student.uclouvain.be wrote:
Hello everyone,
On 06/16/2014 01:31 AM, Akemi Yagi wrote:
Adding patches is absolutely one of the things that the plus kernel offers. In this case, as Guillaume stated, the modifications are quite extensive. Without understanding all the details of the patches, I would like to ask if there are any consequences that might "negatively" affect the existing plus kernel users.
As I am not a developer of Multipath-TCP, I asked the main developer (Christoph Paasch) about that. Here is what he answered (translated from french):
I see no « negative consequences » in the inclusion of MPTCP in the centosplus kernel. Everything should work normally (and if its not, it’s a bug we can fix). The impact on the performance of plain TCP should be very low.
So, we can assume adding the proposed patches should be transparent to the plus kernel users, I suppose. Is this the case?
KB? To go ahead with this or not? A separate specific kernel?
Akemi
On 06/17/2014 06:33 PM, Akemi Yagi wrote:
On Mon, Jun 16, 2014 at 2:41 AM, Guillaume Derval guillaume.derval@student.uclouvain.be wrote:
Hello everyone,
On 06/16/2014 01:31 AM, Akemi Yagi wrote:
Adding patches is absolutely one of the things that the plus kernel offers. In this case, as Guillaume stated, the modifications are quite extensive. Without understanding all the details of the patches, I would like to ask if there are any consequences that might "negatively" affect the existing plus kernel users.
As I am not a developer of Multipath-TCP, I asked the main developer (Christoph Paasch) about that. Here is what he answered (translated from french):
I see no « negative consequences » in the inclusion of MPTCP in the centosplus kernel. Everything should work normally (and if its not, it’s a bug we can fix). The impact on the performance of plain TCP should be very low.
So, we can assume adding the proposed patches should be transparent to the plus kernel users, I suppose. Is this the case?
KB? To go ahead with this or not? A separate specific kernel?
either way you decide, i can help facilitate it either way
On Tue, Jun 17, 2014 at 10:37 AM, Karanbir Singh mail-lists@karan.org wrote:
On 06/17/2014 06:33 PM, Akemi Yagi wrote:
So, we can assume adding the proposed patches should be transparent to the plus kernel users, I suppose. Is this the case?
KB? To go ahead with this or not? A separate specific kernel?
either way you decide, i can help facilitate it either way
With my initial concern gone, I now have more practical question.
The modifications are quite extensive as noted in the beginning. I count 15 patches in include/ and 29 in /net . For each kernel update, the patch set must be updated/tested. Also there may be updates to the patches themselves within a given kernel release. How can this be best handled? Could dealing with the patches cause a delay in the cplus kernel release?
Having a separate kernel addressing the mptcp stuff might work around that potential issue.
Akemi
Le 17 juin 2014 à 19:57, Akemi Yagi amyagi@gmail.com a écrit :
With my initial concern gone, I now have more practical question.
The modifications are quite extensive as noted in the beginning. I count 15 patches in include/ and 29 in /net . For each kernel update, the patch set must be updated/tested. Also there may be updates to the patches themselves within a given kernel release. How can this be best handled? Could dealing with the patches cause a delay in the cplus kernel release?
Having a separate kernel addressing the mptcp stuff might work around that potential issue.
Akemi
I personally think that creating a separate kernel (kernel-cplus-mptcp? :-) ) is the best thing to do.
Best regards, Guillaume Derval.
On Tue, Jun 17, 2014 at 12:51 PM, Guillaume Derval guillaume.derval@student.uclouvain.be wrote:
Le 17 juin 2014 à 19:57, Akemi Yagi amyagi@gmail.com a écrit :
With my initial concern gone, I now have more practical question.
The modifications are quite extensive as noted in the beginning. I count 15 patches in include/ and 29 in /net . For each kernel update, the patch set must be updated/tested. Also there may be updates to the patches themselves within a given kernel release. How can this be best handled? Could dealing with the patches cause a delay in the cplus kernel release?
Having a separate kernel addressing the mptcp stuff might work around that potential issue.
Akemi
I personally think that creating a separate kernel (kernel-cplus-mptcp? :-) ) is the best thing to do.
Best regards, Guillaume Derval.
That probably makes everyone's life easier. And KB says he would provide support either way. In this case, it will be a distro kernel derivative, so a 'kernel-mptcp' ?
Akemi
On 06/17/2014 10:59 PM, Akemi Yagi wrote:
On Tue, Jun 17, 2014 at 12:51 PM, Guillaume Derval guillaume.derval@student.uclouvain.be wrote:
Le 17 juin 2014 à 19:57, Akemi Yagi amyagi@gmail.com a écrit :
With my initial concern gone, I now have more practical question.
The modifications are quite extensive as noted in the beginning. I count 15 patches in include/ and 29 in /net . For each kernel update, the patch set must be updated/tested. Also there may be updates to the patches themselves within a given kernel release. How can this be best handled? Could dealing with the patches cause a delay in the cplus kernel release?
Having a separate kernel addressing the mptcp stuff might work around that potential issue.
Akemi
I personally think that creating a separate kernel (kernel-cplus-mptcp? :-) ) is the best thing to do.
Best regards, Guillaume Derval.
That probably makes everyone's life easier. And KB says he would provide support either way. In this case, it will be a distro kernel derivative, so a 'kernel-mptcp' ?
+1 for separate kernel-mptcp
On 06/17/2014 02:51 PM, Guillaume Derval wrote:
Le 17 juin 2014 à 19:57, Akemi Yagi amyagi@gmail.com a écrit :
With my initial concern gone, I now have more practical question.
The modifications are quite extensive as noted in the beginning. I count 15 patches in include/ and 29 in /net . For each kernel update, the patch set must be updated/tested. Also there may be updates to the patches themselves within a given kernel release. How can this be best handled? Could dealing with the patches cause a delay in the cplus kernel release?
Having a separate kernel addressing the mptcp stuff might work around that potential issue.
Akemi
I personally think that creating a separate kernel (kernel-cplus-mptcp? :-) ) is the best thing to do.
I think a separate kernel might be best too.
It can be kernel-mptcp and live in centos-plus if we choose to do kernel-plus, kernel-<some_name>, etc. or we can do it another way, but regardless in a separate kernel is likely best.
On 06/17/2014 03:00 PM, Johnny Hughes wrote:
On 06/17/2014 02:51 PM, Guillaume Derval wrote:
Le 17 juin 2014 à 19:57, Akemi Yagi amyagi@gmail.com a écrit :
With my initial concern gone, I now have more practical question.
The modifications are quite extensive as noted in the beginning. I count 15 patches in include/ and 29 in /net . For each kernel update, the patch set must be updated/tested. Also there may be updates to the patches themselves within a given kernel release. How can this be best handled? Could dealing with the patches cause a delay in the cplus kernel release?
Having a separate kernel addressing the mptcp stuff might work around that potential issue.
Akemi
I personally think that creating a separate kernel (kernel-cplus-mptcp? :-) ) is the best thing to do.
I think a separate kernel might be best too.
It can be kernel-mptcp and live in centos-plus if we choose to do kernel-plus, kernel-<some_name>, etc. or we can do it another way, but regardless in a separate kernel is likely best.
That also means that someone (or someones) need to volunteer to maintain it as well.
On 06/17/2014 04:01 PM, Johnny Hughes wrote:
On 06/17/2014 03:00 PM, Johnny Hughes wrote:
On 06/17/2014 02:51 PM, Guillaume Derval wrote:
Le 17 juin 2014 à 19:57, Akemi Yagi amyagi@gmail.com a écrit :
With my initial concern gone, I now have more practical question.
The modifications are quite extensive as noted in the beginning. I count 15 patches in include/ and 29 in /net . For each kernel update, the patch set must be updated/tested. Also there may be updates to the patches themselves within a given kernel release. How can this be best handled? Could dealing with the patches cause a delay in the cplus kernel release?
Having a separate kernel addressing the mptcp stuff might work around that potential issue.
Akemi
I personally think that creating a separate kernel (kernel-cplus-mptcp? :-) ) is the best thing to do.
I think a separate kernel might be best too.
It can be kernel-mptcp and live in centos-plus if we choose to do kernel-plus, kernel-<some_name>, etc. or we can do it another way, but regardless in a separate kernel is likely best.
That also means that someone (or someones) need to volunteer to maintain it as well.
Would it be a distro kernel with only the mptcp added or would it be a "plus" kernel with mptcp?
Thanks,
On 06/18/2014 11:46 AM, Steve Clark wrote:
Would it be a distro kernel with only the mptcp added or would it be a "plus" kernel with mptcp?
at the moment, the conversation seems to be going down the route of a mptcp specific kernel, with no direct overlap with plus. Given enough testing and user traction, in the future this might get rolled into the plus kernel
-
Akemi Yagi -
Guillaume Derval -
Johnny Hughes -
Karanbir Singh -
Manuel Wolfshant -
Steve Clark