Differences

This shows you the differences between two versions of the page.

--- 0x13:reports:d1t1t03-traffic-control-workshop [2019/03/27 22:34] – created ehalep
+++ 0x13:reports:d1t1t03-traffic-control-workshop [2019/09/28 17:04] (current) – external edit 127.0.0.1
@@ Line 2: / Line 2: @@
 Workshop: Traffic Control Workshop
 Chair: Jamal Hadi Salim
-Report by:
+Report by: Jamal Hadi Salim
+This workshop is a regular fixture at the netdev conference to discuss ongoing issues, experiences, new features and proposals.
-Site: https://www.netdevconf.org/0x13/session.html?workshop-Traffic-Control
+contributor: Vlad Buslov
-Slides:
+topic:  Unlocking TC rules update API: challenges and lessons learned.
-Videos:
+Vlad Discussed his ongoing work on removing locks from the TC rules control path in order to improve update rates. Vlad hopes that when he is done he can improve the infrastructure by having multiple control path updates happening concurrently. In general Vlad discussed his challenges and lessons he learned while progressing on this work.
+Contributors: Amritha Nambiar, Anjali Singhai and Sridhar Samudrala
+Topic: Intel TC Hardware Offload Updates
+Amritha talked in details about the work done and challenges encountered for creating port range expressions for the Flower classifier. She then proceeded to describe new work done on ixgbe for the u32 classifier, namely new actions created (skbedit on mark and accept/ok) as well as ability to
+define queue selection based on flowid/classid definition.
+Anjali went on to describe future work to introduce stats on a per filter basis based on policy and intention to add policer work. She talked about challenges with Flower having named fields which always require kernel changes; rules being cached at many layers (user space, kernel, hardware) and last described intention to take advantage of Vlad's work optimizing rule update to 500K/sec for Intel hardware.
+Contributor: Andy Gospodarek
+Topic: Stats offload transfers
+Andy discussed the challenges of offloaded TC stats in presence of a large number(millions) of rules/actions; Many use cases (such as billing or SLA validations) require frequent updates from hardware at a higher frequency which are not efficient (true for software as well). Several approaches were discussed. This will be an ongoing discussion.
+Contributor: John Hurley (Presented by Simon Horman)
+Topic: Offloading TC rules on OVS Internal Ports
+Simon described a challenge in regards to semantics used by "OVS Internal Ports"; internal ports send packets directly to the stack and bypass TC ingress(and therefore miss offloading opportunity presented by TC). Several proposals were made to address these issues.
+Contributor: Lucas Bates
+Topic: Testing Developments for TDC
+Lucas described a new tdc feature for traffic generation using scappy. He then proceeded to give
+a demo on its usage. The goal of the feature is to be able to produce arbitrary packets that can test classifier rules.
+Contributor: Roman Mashak
+Roman described the Quota Enforcement(QE) action which is to be upstreamed. QE is used in telco environments to provide a limit on bytes consumed during a policy lifetime (think your monthly 4GB data plan)
+Site: https://www.netdevconf.info/0x13/session.html?workshop-Traffic-Control