[dpdk-dev] [RFC] Add Membership Library

[Stephen Hemminger]

On Thu, 1 Jun 2017 01:03:36 +0000
"Wang, Yipeng1" <yipeng1.wang at intel.com> wrote:

> Hi Vincent,
> 
> Thanks for the comments and some quick responses below:
> 
> - DPDK Bloom Filter is derived from the libbloom (as shown by the included BSD license), but optimized for performance with various DPDK goodness such as rte_zmalloc, rte_bitmap, rte_jhash, etc. It doesn't seem appropriate to    bring those DPDK specific features into libbloom, which is designed for more generic environments.
> - DPDK Bloom Filter is just one very basic mode among other more sophisticated ones in the DPDK Membership Library, which supports a different set of APIs (e.g., bulk lookup, multi-match, etc.) than those supported in libbloom.
> - It may make sense to bring some generic (i.e., non DPDK specific) improvements back to libbloom once identified.
> 
> Thanks
> Yipeng
> 
> > -----Original Message-----
> > From: Vincent Jardin [mailto:vincent.jardin at 6wind.com]
> > Sent: Saturday, May 27, 2017 2:42 AM
> > To: Wang, Yipeng1 <yipeng1.wang at intel.com>
> > Cc: dev at dpdk.org; Gobriel, Sameh <sameh.gobriel at intel.com>; Wang, Ren
> > <ren.wang at intel.com>; Tai, Charlie <charlie.tai at intel.com>
> > Subject: Re: [dpdk-dev] [RFC] Add Membership Library
> > 
> > Why duplicating Jyri's libbloom - https://github.com/jvirkki/libbloom - for this
> > DPDK capability? Why not showing that you can contribute to libbloom and
> > make it linkable with the DPDK?
> > 
> > There are so many duplicated code...
> > 
> > Thank you,
> >   Vincent
> >   
> 

Note: rte_zmalloc is really no faster in my experience than regular malloc. And under
memory pressure rte_malloc is worse than malloc.