[dpdk-dev] [PATCH v2] ethdev: add fuzzy match pattern
Zhang, Qi Z
qi.z.zhang at intel.com
Tue Jun 13 08:03:19 CEST 2017
> -----Original Message-----
> From: Adrien Mazarguil [mailto:adrien.mazarguil at 6wind.com]
> Sent: Monday, June 12, 2017 11:39 PM
> To: Thomas Monjalon <thomas at monjalon.net>
> Cc: Zhang, Qi Z <qi.z.zhang at intel.com>; dev at dpdk.org
> Subject: Re: [dpdk-dev] [PATCH v2] ethdev: add fuzzy match pattern
> On Thu, Jun 08, 2017 at 09:19:24AM +0200, Thomas Monjalon wrote:
> > Hi,
> > 08/06/2017 00:21, Qi Zhang:
> > > Add new meta pattern item RTE_FLOW_TYPE_ITEM_FUZZY.
> > I disagree about fuzzy wording, because fuzzy is something different I
> > think.
> At least it's a generic term, unrelated PMDs could likewise implement
> another kind of fuzzy matching for performance reasons (for applications
> that really care more about performance than accuracy). The fact this
> particular implementation is built on top of something that is internally
> known as signature hash should not be relevant.
> > > This is for device that support fuzzy match option.
> > > Usually a fuzzy match is fast but the cost is accuracy.
> > > i.e. Signature Match only match pattern's hash value, but it is
> > > possible two different patterns have the same hash value.
> > You have described it yourself here: it matches a hash of the signature.
> > Why not using "hash" as wording?
> While "hash" would be also correct, in my opinion it would restrict this
> pattern item to Intel adapters (ixgbe) with the ability to actually perform a
> hash on patterns and use the resulting value to imperfectly match traffic.
> This excludes other implementations with similar unpredictable results.
> Also a threshold notion is necessary as far as I understand, the signature
> mode has several levels. The higher, the fuzzier it gets. This behavior is
> difficult to translate to something generic named "hash", as one would have
> to describe how the hash is computed and packets matched according to
> that value, which really is ixgbe-specific at the moment.
> > > Matching accuracy level can be configure by subfield threshold.
> > > Driver can divide the range of threshold and map to different
> > > accuracy levels that device support.
> > >
> > > Signed-off-by: Qi Zhang <qi.z.zhang at intel.com>
> Qi, many of my other comments about v1 still stand, please check my
> previous reply in any case.
Sorry, seems I skipped that important mail by mistake, just find it, thanks for all the comments.
> Adrien Mazarguil
More information about the dev