Re: [vpp-dev] bihash change

2020-09-30 Thread Damjan Marion via lists.fd.io


> On 30.09.2020., at 17:29, Jon Loeliger via lists.fd.io 
>  wrote:
> 
> On Wed, Sep 30, 2020 at 8:11 AM Damjan Marion via lists.fd.io 
>   > wrote:
> 
> Just a heads-up on under-the-hood change[1] which is merged few minutes ago 
> and which affects all features that use bihash.
> Now, bihash is allocating memory from the (main) heap, instead of mmap()-ing 
> it’s own alloc arena.
> 
> [ ...] 
> Damjan
> 
> Damjan,
> 
> Does this include the bihash and heap use within the Classifier code as well?
> Or just the templated bihash from the library?

no, no changes for classifier….


-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#17619): https://lists.fd.io/g/vpp-dev/message/17619
Mute This Topic: https://lists.fd.io/mt/77216768/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-



Re: [vpp-dev] bihash change

2020-09-30 Thread Jon Loeliger via lists.fd.io
On Wed, Sep 30, 2020 at 8:11 AM Damjan Marion via lists.fd.io  wrote:

>
> Just a heads-up on under-the-hood change[1] which is merged few minutes
> ago and which affects all features that use bihash.
> Now, bihash is allocating memory from the (main) heap, instead of
> mmap()-ing it’s own alloc arena.
>
> [ ...]
> Damjan
>

Damjan,

Does this include the bihash and heap use within the Classifier code as
well?
Or just the templated bihash from the library?

Thanks,
jdl

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#17617): https://lists.fd.io/g/vpp-dev/message/17617
Mute This Topic: https://lists.fd.io/mt/77216768/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-



[vpp-dev] bihash change

2020-09-30 Thread Damjan Marion via lists.fd.io

Guys,

Just a heads-up on under-the-hood change[1] which is merged few minutes ago and 
which affects all features that use bihash.

Now, bihash is allocating memory from the (main) heap, instead of mmap()-ing 
it’s own alloc arena.

There are several beneifs of that approach:
 - smaller memory footprint
 - ability to use hugepages without dedicating them to specific feature
 - smaller coredump files
 
Size of each table is now limited by heap size, and not by the value provided 
during the table initialisation.
In some cases increasing heap size may be needed, and if performance matters 
putting the main heap into hugepages may also be a good idea, as all bihash 
tables will now immediately benefit out of it.

i.e. startup.conf example:

memory {
main-heap-size 2G
main-heap-page-size 1G
}

If anybody notices some issues, please let us know.

— 
Damjan


[1] https://gerrit.fd.io/r/c/vpp/+/29099



-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#17616): https://lists.fd.io/g/vpp-dev/message/17616
Mute This Topic: https://lists.fd.io/mt/77216768/21656
Group Owner: vpp-dev+ow...@lists.fd.io
Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-