I can tell no FIB

I can tell no FIB

[Scott Feldman]

Hi Alex, turns out you're required to take a mandatory week-long
vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
week!  Take lots of pictures.

-scott

Re: I can tell no FIB

[Dave Taht]

On Thu, Mar 5, 2015 at 12:49 PM, Scott Feldman <sfeldma@gmail.com> wrote:
> Hi Alex, turns out you're required to take a mandatory week-long
> vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
> week!  Take lots of pictures.

+10!

Is anyone here working on testing the new FIB stuff on itty bitty 32
bit platforms? It looks really promising
but openwrt is stabilizing on 3.18 and the prospect of backporting all
this stuff to that to test at scale is intimidating. (*I* am willing
to wait for 4.2)

But boy, could openwrt test at scale:

https://downloads.openwrt.org/snapshots/trunk/

>
> -scott
> --
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html



-- 
Dave Täht
Let's make wifi fast, less jittery and reliable again!

https://plus.google.com/u/0/107942175615993706558/posts/TVX3o84jjmb

Re: I can tell no FIB

[Alexander Duyck]

On 03/05/2015 12:49 PM, Scott Feldman wrote:
> Hi Alex, turns out you're required to take a mandatory week-long
> vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
> week!  Take lots of pictures.
>
> -scott

Well I have one more set of 9 and then I will stop for a while.  I was
planning to send it later tonight.  Then I can probably take that
week-long vacation..

I'm going to drop the portion of the patches I had where I was
up-levelling the key vector since I still don't have a solution for the
extra costs of insertion/deletion from the trie.  Also I don't think it
would work well with the merge of the main and local tries if that is
the route we are planning to take.

- Alex

Re: I can tell no FIB

[David Miller]

From: Alexander Duyck <alexander.duyck@gmail.com>
Date: Thu, 05 Mar 2015 13:17:54 -0800

> On 03/05/2015 12:49 PM, Scott Feldman wrote:
>> Hi Alex, turns out you're required to take a mandatory week-long
>> vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
>> week!  Take lots of pictures.
>>
>> -scott
> 
> Well I have one more set of 9 and then I will stop for a while.  I was
> planning to send it later tonight.  Then I can probably take that
> week-long vacation..
> 
> I'm going to drop the portion of the patches I had where I was
> up-levelling the key vector since I still don't have a solution for the
> extra costs of insertion/deletion from the trie.  Also I don't think it
> would work well with the merge of the main and local tries if that is
> the route we are planning to take.

Ok, where do you want to place the main/local tree patch then?

Scott's L3 work logically depends upon that, and actually I think
Scott is sending you on vacation so that he doesn't have to rebase so
much :-)

Re: I can tell no FIB

[Alexander Duyck]

On 03/05/2015 01:27 PM, David Miller wrote:
> From: Alexander Duyck <alexander.duyck@gmail.com>
> Date: Thu, 05 Mar 2015 13:17:54 -0800
>
>> On 03/05/2015 12:49 PM, Scott Feldman wrote:
>>> Hi Alex, turns out you're required to take a mandatory week-long
>>> vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
>>> week!  Take lots of pictures.
>>>
>>> -scott
>> Well I have one more set of 9 and then I will stop for a while.  I was
>> planning to send it later tonight.  Then I can probably take that
>> week-long vacation..
>>
>> I'm going to drop the portion of the patches I had where I was
>> up-levelling the key vector since I still don't have a solution for the
>> extra costs of insertion/deletion from the trie.  Also I don't think it
>> would work well with the merge of the main and local tries if that is
>> the route we are planning to take.
> Ok, where do you want to place the main/local tree patch then?
>
> Scott's L3 work logically depends upon that, and actually I think
> Scott is sending you on vacation so that he doesn't have to rebase so
> much :-)

Yeah, I kind of figured that might be the case.  If needed I can hold
off for a day or so while Scott gets the FIB offloading stuff in and I
could just submit the remaining 9 plus the main/local merge patch as an
RFC so that it can be reviewed while the offload stuff is accepted.

- Alex

Re: I can tell no FIB

[Alexander Duyck]

On 03/05/2015 01:01 PM, Dave Taht wrote:
> On Thu, Mar 5, 2015 at 12:49 PM, Scott Feldman <sfeldma@gmail.com> wrote:
>> Hi Alex, turns out you're required to take a mandatory week-long
>> vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
>> week!  Take lots of pictures.
> +10!
>
> Is anyone here working on testing the new FIB stuff on itty bitty 32
> bit platforms? It looks really promising
> but openwrt is stabilizing on 3.18 and the prospect of backporting all
> this stuff to that to test at scale is intimidating. (*I* am willing
> to wait for 4.2)
>
> But boy, could openwrt test at scale:
>
> https://downloads.openwrt.org/snapshots/trunk/

As far as what to backport I would recommend only targeting the FIB 
changes that went into 4.0.  It was low risk for causing regressions and 
significant benefit.  This is why I referred to them as "low hanging 
fruit" when I described them  in my presentation at Netdev 0.1.  The 
stuff that went into 4.0 reduced things by hundreds of nanosecnds in 
some cases, the stuff targeting net-next/4.1 is only going to reduce 
things by tens of nanoseconds.

What I am working on now is basically just trying squeeze the last bits 
of performance out of what is left.    Excluding the main/local merge I 
have been able to remove 20% (10 - 35ns depending on the test) of the 
remaining CPU overhead for the fib table look-up with what has been 
submitted since net-next reopened.  The local/main trie merge pushes 
that to somewhere around a 40% reduction from what I have seen.

- Alex

Re: I can tell no FIB

[Scott Feldman]

On Thu, Mar 5, 2015 at 2:01 PM, Alexander Duyck
<alexander.duyck@gmail.com> wrote:
> On 03/05/2015 01:27 PM, David Miller wrote:
>> From: Alexander Duyck <alexander.duyck@gmail.com>
>> Date: Thu, 05 Mar 2015 13:17:54 -0800
>>
>>> On 03/05/2015 12:49 PM, Scott Feldman wrote:
>>>> Hi Alex, turns out you're required to take a mandatory week-long
>>>> vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
>>>> week!  Take lots of pictures.
>>>>
>>>> -scott
>>> Well I have one more set of 9 and then I will stop for a while.  I was
>>> planning to send it later tonight.  Then I can probably take that
>>> week-long vacation..
>>>
>>> I'm going to drop the portion of the patches I had where I was
>>> up-levelling the key vector since I still don't have a solution for the
>>> extra costs of insertion/deletion from the trie.  Also I don't think it
>>> would work well with the merge of the main and local tries if that is
>>> the route we are planning to take.
>> Ok, where do you want to place the main/local tree patch then?
>>
>> Scott's L3 work logically depends upon that, and actually I think
>> Scott is sending you on vacation so that he doesn't have to rebase so
>> much :-)
>
> Yeah, I kind of figured that might be the case.  If needed I can hold
> off for a day or so while Scott gets the FIB offloading stuff in and I
> could just submit the remaining 9 plus the main/local merge patch as an
> RFC so that it can be reviewed while the offload stuff is accepted.

Holding off for a day or two would help; I think my next v4 set I'm
testing now and sending later today will stick.

I think I've had to rebase my FIB changes 3 times so far, and it's
getting harder each time because more is moving around as you peel the
onion.  But your work is awesome, so don't take it the wrong way.

I'll need you to review my v4 changes in fib_trie.c.  I had to clone
your new fib_table_flush() to make one to clear out external FIB
entries (those FIB entries that where offloaded externally).  I'm not
happy with the duplication of code, but I wasn't seeing a clean way to
break it apart.  You have a goto jumping back into the middle of a
while loop, and my tiny brain can't deal with that.  :)  I'm hoping
you can look at it and see a way to minimize duplicate code.

Thanks Alex.

-scott

Re: I can tell no FIB

[Alexander Duyck]

On 03/05/2015 04:07 PM, Scott Feldman wrote:
> On Thu, Mar 5, 2015 at 2:01 PM, Alexander Duyck
> <alexander.duyck@gmail.com> wrote:
>> On 03/05/2015 01:27 PM, David Miller wrote:
>>> From: Alexander Duyck <alexander.duyck@gmail.com>
>>> Date: Thu, 05 Mar 2015 13:17:54 -0800
>>>
>>>> On 03/05/2015 12:49 PM, Scott Feldman wrote:
>>>>> Hi Alex, turns out you're required to take a mandatory week-long
>>>>> vacation after your fourth patch set to net/ipv4/fib_*.  See you in a
>>>>> week!  Take lots of pictures.
>>>>>
>>>>> -scott
>>>> Well I have one more set of 9 and then I will stop for a while.  I was
>>>> planning to send it later tonight.  Then I can probably take that
>>>> week-long vacation..
>>>>
>>>> I'm going to drop the portion of the patches I had where I was
>>>> up-levelling the key vector since I still don't have a solution for the
>>>> extra costs of insertion/deletion from the trie.  Also I don't think it
>>>> would work well with the merge of the main and local tries if that is
>>>> the route we are planning to take.
>>> Ok, where do you want to place the main/local tree patch then?
>>>
>>> Scott's L3 work logically depends upon that, and actually I think
>>> Scott is sending you on vacation so that he doesn't have to rebase so
>>> much :-)
>> Yeah, I kind of figured that might be the case.  If needed I can hold
>> off for a day or so while Scott gets the FIB offloading stuff in and I
>> could just submit the remaining 9 plus the main/local merge patch as an
>> RFC so that it can be reviewed while the offload stuff is accepted.
> Holding off for a day or two would help; I think my next v4 set I'm
> testing now and sending later today will stick.

Well the last batch of patches are out there, but there isn't any rush
on them.  If need be Dave can just leave them in review until we get the
thumbs up or down on the v4 changes since I think you and I are probably
the only two patching that region for now.  I suspect the changes
shouldn't be as bad as the past patches have been since most of it is
just reworking structures and updating the resize code.

> I think I've had to rebase my FIB changes 3 times so far, and it's
> getting harder each time because more is moving around as you peel the
> onion.  But your work is awesome, so don't take it the wrong way.

Yeah, a lot of it was pretty messy since I was having to untangle some
messed up code.

> I'll need you to review my v4 changes in fib_trie.c.  I had to clone
> your new fib_table_flush() to make one to clear out external FIB
> entries (those FIB entries that where offloaded externally).  I'm not
> happy with the duplication of code, but I wasn't seeing a clean way to
> break it apart.  You have a goto jumping back into the middle of a
> while loop, and my tiny brain can't deal with that.  :)  I'm hoping
> you can look at it and see a way to minimize duplicate code.

Just CC me on the patches and I will look them over when you submit
them.  You may want to consider just adding a flag field to
fib_table_flush call since the determining factor for dropping a
fib_alias is based on if the fib_info has the RTNH_F_DEAD flag set, you
could pass that flag to the legacy callers for fib_table_flush and pass
RTNH_F_EXTERNAL for your new call.

> Thanks Alex.
>
> -scott

No problem, as is things for me are winding down in terms of the FIB. 
I'm just flushing my backlog at this point so as long as I can get the
last 9 flushed out before net-next closes I'll be good.

- Alex

Re: I can tell no FIB

[David Miller]

From: Alexander Duyck <alexander.duyck@gmail.com>
Date: Thu, 05 Mar 2015 16:20:43 -0800

> Well the last batch of patches are out there, but there isn't any rush
> on them.  If need be Dave can just leave them in review until we get the
> thumbs up or down on the v4 changes since I think you and I are probably
> the only two patching that region for now.  I suspect the changes
> shouldn't be as bad as the past patches have been since most of it is
> just reworking structures and updating the resize code.

Scott just tell me what you want me to do, I can hold off on
Alex's current series while you put v4 of your stuff together.

Re: I can tell no FIB

[Scott Feldman]

On Thu, Mar 5, 2015 at 7:20 PM, David Miller <davem@davemloft.net> wrote:
> From: Alexander Duyck <alexander.duyck@gmail.com>
> Date: Thu, 05 Mar 2015 16:20:43 -0800
>
>> Well the last batch of patches are out there, but there isn't any rush
>> on them.  If need be Dave can just leave them in review until we get the
>> thumbs up or down on the v4 changes since I think you and I are probably
>> the only two patching that region for now.  I suspect the changes
>> shouldn't be as bad as the past patches have been since most of it is
>> just reworking structures and updating the resize code.
>
> Scott just tell me what you want me to do, I can hold off on
> Alex's current series while you put v4 of your stuff together.

I just sent v4.  I need Alex to look over my fib changes.  My testing
didn't hit any problems, but lots of things are moving right where I
was making some insertions and I want to make sure it's clean from a
coding standpoint and also clean from a calling-context standpoint.

Also Alex, can you run my v4 patches on your benchmark setup?  You'll
not have rocker so nothing is getting offloaded, but I wanted to see
if your benchmarks show any degradation from your baseline, for route
insertions and deletions.  There is a per-route cost associated with
my changes, but I'm hoping the costs are negligible.  Please run two
tests: one with CONFIG_NET_SWITCHDEV=y and one with
CONFIG_NET_SWITCHDEV=n.  Thanks Alex!

And thanks Dave for pushing me along on v4.

-scott