We read every piece of feedback, and take your input very seriously.
To see all available qualifiers, see our documentation.
Quality-of-Service: Mention headroom doubling on 8-lane ports Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Describe descriptor buffers Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Add Spectrum-3 to default shared buffer configuration Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Fix a typo, "dcb etc" -> "dcb ets" Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Clarify TC/DCB mode behavior Signed-off-by: Ido Schimmel <[email protected]> Reviewed-by: Petr Machata <[email protected]>
Add a RoCE howto The howto is a direct translation of an Onyx-specific guide, with adjustments to make it up to date, and to make it portable to Spectrum-2+ (notably the pool sizes). The guide uses dynamic pool for the lossless pool. In the QoS page of the wiki, we currently recommend to always use static pools, but it turns out this is not necessary, and dynamic pool will work just as well. Therefore, strike the requirement that the lossless pool be static from the QoS page. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Fix a typo in dcb invocation Reported-by: Maksym Yaremchuk <[email protected]> Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Convert to iproute2 dcb, drop mlnx_qos The dcb tool is now available in the released version of iproute2, and can be taken as the primary tool to configure DCB in Linux. Drop all mentions of mlnx_qos and convert to iproute2 dcb. In the text, always mention iproute2 dcb before lldpad. Change also some exposition examples from lldptool to iproute2 dcb. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Explain PG buffer overallocation mlxsw will apparently overallocate due to rounding and other concerns. Describe that in the wiki. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Describe manual PG buffer configuration Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Describe mlnx_qos usage We will have to document mlnx_qos for the manual PG configuration anyway. Might as well do it systematically for all the topics. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Add a section on Priority Group Buffers The content is currently very sketchy, but will grow once the DCB and TC modes are explained and mlnx_qos brought in. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Convert DCB section to Traffic Scheduling This section now holds exclusively content on the topic of traffic scheduling. Rename it appropriately and reorganize bits of it. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Rework the Losslessness section This section needs to explain the Xoff threshold more prominently, because we will need to refer to it as a term when describing the minimum PG size and delay allowance for manually configured PGs. We can also separate out the bits that are related purely to LLDP configuration from the bits that actually configure the HW. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Consolidate discussion of losslessness The fact that lossless flows are configured through DCB is not that fundamental. Add a new section specifically for the topic of configuring lossless data flows. Move there the PFC stuff from DCB, and PAUSE. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Add a chapter about tools This will later discuss mlnx_qos as well. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Move DSCP rewrite This happens later in the pipeline and therefore should be discussed further down. I still place it before shared buffers, because those feel special enough to be a "tack-on", like the CoPP. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Move max delay remark The remark about maximum delay relates specifically to the case that PG buffers are autoconfigured. Move it upwards toward the command that handles it. Also change the reference to "bit times" to "bit". Bit times are used for PFC pause frames requesting sender pause. The delay provision is simply in bits. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Move first-app-rule paragraph This holds generally, whatever the way the first APP rule is added. Move it up so that it later covers mlnx_qos as well. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Move and extend notes about SKB priority Mentioning SKB priority makes better sense in the context of priority assignment than ETS. Move the paragraph accordingly. The SKB priority is analogous to switch priority, but is not the same. Make it clear that SP is not projected to slow path. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Tweak default priority Show both how to add a default priority rule and how to delete it, similarly to how other APP rules have it. Likewise add explanatory comments. Move the explanation of how ambiguities are resolved up, because it is a general principle not related to particular syntax. Adjust wording to make it clear the Linux actually allows duplicate rules, not Open LLDP specifically. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Illustrate how PAUSE may not be implemntable Put more words towards explaining the PAUSE dealy constant and how it can be too large if too many buffers are used with too-large MTU. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Fix a typo, buffers->pools The memory areas where traffic waits for transmission are called "pools". Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Redirect open-lldp.org references elsewhere open-lldp.org is down. Redirect the main reference to github, and the DCB overview article reference to web.archive.org. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Don't call it The Packet's Path This sounds vaguely samurai-like and probably doesn't convey that the section talks not about morals and ethos, but cold algorithmical rules. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Normalize command line prompts The wiki pages use mostly $ as a prompt. Use it on this page as well. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Normalize switch port references The names of switch ports are canonically swp<X>, not sw<X>p<Y>. Simply change all example port names to swp1. Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Reformat the features-by-version table Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>
Quality-of-Service: Describe Control Plane Policing (CoPP) Signed-off-by: Ido Schimmel <[email protected]> Reviewed-by: Petr Machata <[email protected]>
Quality-of-Service, ACLs: Describe ACL-based DSCP rewrite Signed-off-by: Petr Machata <[email protected]> Signed-off-by: Ido Schimmel <[email protected]>