netdev - Re: network stream fairness

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <1448818812.24696.104.camel@edumazet-glaptop2.roam.corp.google.com>
Date:	Sun, 29 Nov 2015 09:40:12 -0800
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Niklas Cassel <niklas.cassel@...s.com>
Cc:	"netdev@...r.kernel.org" <netdev@...r.kernel.org>
Subject: Re: network stream fairness

> If I didn't know that though, my expectations would be that the
> Fair Queue packet scheduler would behave just like its name says, fair. :P
> 

Only if the flows have same characteristics. This cannot be ensured for
hours long flows. This should be obvious.

If one flow is not trying to push enough, it will loose its round in the
Round Robin scheduler, because no packet is sitting in the packet
scheduler for this flow.

A simple slowdown, cause by a packet drop, or scheduler artifact on the
sender or receiver can make one flow steal whole bandwidth, even when FQ
is in the picture.

I have no idea why you are expecting multiple TCP flows being fair.
There is absolutely nothing in TCP protocol to ensure this.

In short, fq_codel or fq are not doing what you think they do.

If you want fair flows, you MUST pace them say at 40Mbits each.

The fact that one 'patch' slightly changes the condition for unfairness
to start is absolutely irrelevant.

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html