Yusheng Wang <yshw...@vmware.com> wrote on 06/17/2016 04:30:02 AM:
> From: Yusheng Wang <yshw...@vmware.com>
> To: Hui Kang/Watson/IBM@IBMUS
> Cc: "dev@openvswitch.org" <dev@openvswitch.org>
> Date: 06/17/2016 04:40 AM
> Subject: Re: [ovs-dev] [PATCH] OVN: initial patch of datalog engine
>
>
> The idea is to write ovn-northd in nlog program. In order to do that,
> we need a hook to ovs-db, which is the next step I am going to do.
> Generally, a subset of ovs-db tables will be input tables to the nlog
> engine and the engine's output table will also be written to ovs-db.
>
> If we think about ovn-northd as a data converter (function) between
> north bound tables and south bound tables, it is straightforward that
> any engine that works in this way could be used.
Thanks for your detailed explanation. This plan is more ambition than I
thought; it involves rewriting the ovn-northd. Look forward to that.
>
> The improvement comes from incremental computation. As for the
> scalability, the paper [2] may not apply to OVN since they are of
> totally different design.
The paper explicitly says the nlog implements the incremental computation
for any state change in the input table, "incremental computation allows us
to recompute only the affected state and push the delta down to the network
edge." Therefore, IMO your implementation is applicable to either OVN or
the NVP [2]. Please correct me if I am wrong.
>
> In order to test the engine, we could try the interactive test case.
There
> is one combo test case in the patch and it has an interactive mode. We
> could load a nlog program and run data on it.
I will try the interactive test case in the patch. One minor thing: do you
mind
removing some trailing whitespace in these patches [5] [6] so that patching
is
easier. Thanks.
Some error message is as follows:
Applying: OVN: datalog man page
/root/Downloads/ovs/.git/rebase-apply/patch:34: trailing whitespace.
</pre>
/root/Downloads/ovs/.git/rebase-apply/patch:58: trailing whitespace.
specify the external function of the engine.
/root/Downloads/ovs/.git/rebase-apply/patch:291: trailing whitespace.
u(a,b) +---+---+ +---+---+ +---+---+---+---+
/root/Downloads/ovs/.git/rebase-apply/patch:293: trailing whitespace.
+---+---+ +---+---+ +---+---+---+---+
/root/Downloads/ovs/.git/rebase-apply/patch:295: trailing whitespace.
+---+---+ +---+---+---+---+
error: patch failed: ovn/lib/automake.mk:43
error: ovn/lib/automake.mk: patch does not apply
[5] https://patchwork.ozlabs.org/patch/623323/
[6] https://patchwork.ozlabs.org/patch/636324/
- Hui
>
> -- Yusheng
> ________________________________________
> From: Hui Kang <ka...@us.ibm.com>
> Sent: Friday, June 17, 2016 3:28 PM
> To: Yusheng Wang
> Cc: dev@openvswitch.org
> Subject: Re: [ovs-dev] [PATCH] OVN: initial patch of datalog engine
>
> Hi, Yusheng,
> I am very intersted in testing your nice nlog implementation. Meanwhile,
I
> appreciate if you could look at some questions I have about this nlog.
>
> - Reading the early proposal of nlog [1], I understand that nlog will
> be serving as a new computation model for ovn-northd. Furthermore, Ithink
the
> plan is to replace the existing join_XXX operations (e.g., [3][4]) in
> populating tables in southbound database. Is my understanding correct?
>
> - How much performance improvement will you expect from using nlog?
> The NVP paper [2] mentioned in section 8.2 that even with nlog, the
> scalability issue still exists because Openflow in hypervisors requires
> O(N^2).
>
> Thanks in advance.
>
> - Hui
>
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
