[#9] Kernel panic with SQM scripts - Comment added

[lede-bugs at lists.infradead.org]

Hi Ansuel,

> On Jun 23, 2016, at 14:58 , lede-bugs at lists.infradead.org wrote:
> 
> The following task has a new comment added:
> 
> FS#9 - Kernel panic with SQM scripts
> User who did this - Ansuel (Ansuel)
> 
> ----------
> thx for the explanation. Currently i’m using the triple wan script with cake 

	Unless you have no NAT or are using mostly IPv6, I predict that this does not work any different from piece_of_cake.qos from the standard sqm-scripts (it might actually consume more CPU while doing the same flow isolation)

> My connetion is a pppoe with atm overhead so is it wrong how i set the sqm settings?

	Hard to say, if you would like to share your /etc/config/sqm I might be able to tell you more, but I am unsure whether this here is the right venue?

> 
> And you did'nt explain the wan variants. It’s the same?

	Okay, I guess I need to go into the details here. The main difference between sqm on WAN and sqm on LAN is related to the fact that the bandwidth directions in the GUI are always from the view of the device sqm is running on (your router). 
	So for the typical WAN situation the download direction for SQM and for internal hosts are aligned, and so are the upload directions. But if you instantiate sqm on a LAN interface the upload from the LAN interface or its egress direction now points into your LAN, so the GUI’s upload field now actually controls the download direction as seen from your hosts. The same apparent directionality flip also holds true for the other direction. 
	The _LAN_ scripts handle that for you by flipping the bandwidth internally so the GUI keeps reflecting the view from end-hosts to the internet. That in itself is not too complicated and could easily be explained to advanced users as yourself (I believe the help texts of the extra scripts should mention something to that regard). But for the dual isolation modes in cake we have a similar flip in the directions, and that gets quite tedious to explain. Triple-isolate is un-directional in that it attempts to do the right thing independent of direction. But triple-isolate is also not fully tested so we need the dual isolation options to sanity check triple-isolate mode
	The dual-dsthost and dual-srchost options however are directional and need to be set correctly, otherwise you might end up with per-external-host-IP fairness instead of the typically expected/requested per-internal-host-IP fairness. For sqm on WAN you want dual-dsthost on ingress and dual-srchost on egress, but for sqm on LAN you want dual-srchost for ingress (since that corresponds to the internal address) and dual-dsthost on egress. 
	And as I did not aim to troubleshoot each and every tester “by hand” (as that has a scalability issue) I opted for creating separate LAN and WAN variants… But please note for most users the LAN versions will not be good solutions since they most likely will either not shape WLAN at all, or worse will apply the shaper to both traffic from LAN to WAN and LAN to WLAN, so talking to a WLAN host will eat upo your internet speed. Have I mentioned that the sqm-scripts-extra package is only for wider testing purposes yet? If not let me repeat they are not for production, but rather to help experimenting to find the best settings to inplement in the sqm-scripts bread-and-butter scripts…
	
Best Regards
        M.

> ----------
> 
> More information can be found at the following URL:
> https://bugs.lede-project.org/index.php?do=details&task_id=9#comment54
> 
> _______________________________________________
> lede-bugs mailing list
> lede-bugs at lists.infradead.org
> http://lists.infradead.org/mailman/listinfo/lede-bugs