Mainstream progress? [message #28275] |
Thu, 13 March 2008 02:20 |
Carl-Daniel Hailfinge
Messages: 15 Registered: February 2007
|
Junior Member |
|
|
Hi,
the ongoing merge of OpenVZ-based lightweight virtualization into
mainline Linux seems to be progressing nicely and I have a few questions
about it:
- How usable will net namespaces be in 2.6.25? I'm especially interested
in assigning a real hardware ethernet device to a VE.
- Will there be an OpenVZ patch or against 2.6.25?
- Do you expect net namespace/virtualization work to be completely
merged in 2.6.26?
- Which OpenVZ features will not make it to mainline in the 2.6.26
timeframe?
Some of the questions above probably can't be answered without guessing,
but a good guess is better than no information and I'd appreciate any
status update newer than the OpenVZ blog entry from 2008-01-26.
Thanks,
Carl-Daniel
|
|
|
Re: Mainstream progress? [message #28299 is a reply to message #28275] |
Thu, 13 March 2008 12:12 |
|
Pavel,
Please correct/enlarge me if there's anything missing/incorrect.
Carl-Daniel Hailfinger wrote:
> Hi,
>
> the ongoing merge of OpenVZ-based lightweight virtualization into
> mainline Linux seems to be progressing nicely and I have a few questions
> about it:
> - How usable will net namespaces be in 2.6.25? I'm especially interested
> in assigning a real hardware ethernet device to a VE.
>
I guess some network namespace stuff is NOT completed in upcoming 2.6.25
(a lot of stuff done, but not finished)
For 2.6.26, it will be more-or-less working, with just a few exceptions
such as conntracks and IPv6 -- maybe.
> - Will there be an OpenVZ patch or against 2.6.25?
>
As of now we do not plan to have one. Porting to each mainstream version
is too much work to do -- instead we concentrate on
(1) merging stuff to mainstream
(2) supporting our stable kernels
(3) stabilizing 2.6.24 (for now)
> - Do you expect net namespace/virtualization work to be completely
> merged in 2.6.26?
>
As said before, we guess that network namespaces will be more or less
useable in 2.6.26.
> - Which OpenVZ features will not make it to mainline in the 2.6.26
> timeframe?
>
A lot. Network namespaces is just one piece of a pie.
To paint in big strokes: resource management and live migration are
still on our todo list.
Resource management: some stuff is in -mm kernel, need to add much more.
Will work on that after finishing with net namespaces.
Live migration: this is much more far perspective, we haven't even
started to discuss that seriously on containers@ list.
> Some of the questions above probably can't be answered without guessing,
> but a good guess is better than no information and I'd appreciate any
> status update newer than the OpenVZ blog entry from 2008-01-26.
>
> Thanks,
> Carl-Daniel
|
|
|
Re: Mainstream progress? [message #28305 is a reply to message #28299] |
Thu, 13 March 2008 13:20 |
Pavel Emelianov
Messages: 1149 Registered: September 2006
|
Senior Member |
|
|
Kir Kolyshkin wrote:
> Pavel,
>
> Please correct/enlarge me if there's anything missing/incorrect.
>
> Carl-Daniel Hailfinger wrote:
>> Hi,
>>
>> the ongoing merge of OpenVZ-based lightweight virtualization into
>> mainline Linux seems to be progressing nicely and I have a few questions
>> about it:
>> - How usable will net namespaces be in 2.6.25? I'm especially interested
>> in assigning a real hardware ethernet device to a VE.
>>
>
> I guess some network namespace stuff is NOT completed in upcoming 2.6.25
> (a lot of stuff done, but not finished)
>
> For 2.6.26, it will be more-or-less working, with just a few exceptions
> such as conntracks and IPv6 -- maybe.
>
>> - Will there be an OpenVZ patch or against 2.6.25?
>>
>
> As of now we do not plan to have one. Porting to each mainstream version
> is too much work to do -- instead we concentrate on
> (1) merging stuff to mainstream
> (2) supporting our stable kernels
> (3) stabilizing 2.6.24 (for now)
2.6.26 will be the next kernel we will port on.
>> - Do you expect net namespace/virtualization work to be completely
>> merged in 2.6.26?
>>
>
> As said before, we guess that network namespaces will be more or less
> useable in 2.6.26.
>
>> - Which OpenVZ features will not make it to mainline in the 2.6.26
>> timeframe?
>>
>
> A lot. Network namespaces is just one piece of a pie.
>
> To paint in big strokes: resource management and live migration are
> still on our todo list.
>
> Resource management: some stuff is in -mm kernel, need to add much more.
> Will work on that after finishing with net namespaces.
Memory management will change with 2.6.26. API will be kept, but
we'll have a per-ve swapout and (probably) drop privvmpages accounting.
> Live migration: this is much more far perspective, we haven't even
> started to discuss that seriously on containers@ list.
>> Some of the questions above probably can't be answered without guessing,
>> but a good guess is better than no information and I'd appreciate any
>> status update newer than the OpenVZ blog entry from 2008-01-26.
>>
>> Thanks,
>> Carl-Daniel
|
|
|
Re: Mainstream progress? [message #30707 is a reply to message #28305] |
Wed, 04 June 2008 13:32 |
Carl-Daniel Hailfinge
Messages: 15 Registered: February 2007
|
Junior Member |
|
|
On 13.03.2008 14:20, Pavel Emelyanov wrote:
> Kir Kolyshkin wrote:
>
>> Pavel,
>>
>> Please correct/enlarge me if there's anything missing/incorrect.
>>
>> Carl-Daniel Hailfinger wrote:
>>
>>> Hi,
>>>
>>> the ongoing merge of OpenVZ-based lightweight virtualization into
>>> mainline Linux seems to be progressing nicely and I have a few questions
>>> about it:
>>> - How usable will net namespaces be in 2.6.25? I'm especially interested
>>> in assigning a real hardware ethernet device to a VE.
>>>
>>>
>> I guess some network namespace stuff is NOT completed in upcoming 2.6.25
>> (a lot of stuff done, but not finished)
>>
>> For 2.6.26, it will be more-or-less working, with just a few exceptions
>> such as conntracks and IPv6 -- maybe.
>>
Do you have an overview of the current 2.6.26 state?
>>> - Will there be an OpenVZ patch or against 2.6.25?
>>>
>>>
>> As of now we do not plan to have one. Porting to each mainstream version
>> is too much work to do -- instead we concentrate on
>> (1) merging stuff to mainstream
>> (2) supporting our stable kernels
>> (3) stabilizing 2.6.24 (for now)
>>
>
> 2.6.26 will be the next kernel we will port on.
>
Yes, I saw the new tree on git.openvz.org. I hope to use a 2.6.26-based
OpenVZ kernel in the near future when 2.6.26 is officially released.
>>> - Do you expect net namespace/virtualization work to be completely
>>> merged in 2.6.26?
>>>
>>>
>> As said before, we guess that network namespaces will be more or less
>> useable in 2.6.26.
>>
It would be cool if you can give some pointers on how to use the merged
part of network namespaces (with OpenVZ tools or any other tool).
>>> - Which OpenVZ features will not make it to mainline in the 2.6.26
>>> timeframe?
>>>
>>>
>> A lot. Network namespaces is just one piece of a pie.
>>
>> To paint in big strokes: resource management and live migration are
>> still on our todo list.
>>
>> Resource management: some stuff is in -mm kernel, need to add much more.
>> Will work on that after finishing with net namespaces.
>>
>
> Memory management will change with 2.6.26. API will be kept, but
> we'll have a per-ve swapout and (probably) drop privvmpages accounting.
>
>
>> Live migration: this is much more far perspective, we haven't even
>> started to discuss that seriously on containers@ list.
>>
That's unfortunate, but I can understand concentrating on a few features
to get them merged completely.
Regards,
Carl-Daniel
|
|
|