netdev - Re: Is 3.18 patch "The tcp: TSO packets automatic sizing" will cause to Troughput drop when enable NO_HZ_IDEL or HIGH_RES

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

Message-ID: <1399393821.15399.12.camel@edumazet-glaptop2.roam.corp.google.com>
Date:	Tue, 06 May 2014 09:30:21 -0700
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	V JobNickname <workofv@...il.com>
Cc:	netdev@...r.kernel.org
Subject: Re: Is 3.18 patch "The tcp: TSO packets automatic sizing" will
 cause to Troughput drop when enable NO_HZ_IDEL or HIGH_RES_TIMERS ?

On Wed, 2014-05-07 at 00:17 +0800, V JobNickname wrote:
> When I porting my ARM platform to Kernel 3.10.0 the network throughput
> of TCP tx is about 160 Mbps.
> 
> After I applied patch-3.10.37, the TCP tx drop to 8x Mbps.
> I found that after I disabled CONFIG_HIGH_RES_TIMERS and NO_HZ_IDEL,
> the throughput can back to 16x Mbps.
> 
> I try to find the more step patches and tile patch-3.10.17, the
> throughput with enabled NO_HZ_IDEL or HIGH_RES_TIMERS can still have
> 16x Mbps
> After applied patch-3.10.17-18, it start to drop the performance.
> 
> More detail trace, the first commit of 3.10.18, tcp: TSO packets
> automatic sizing, cause to this issue.
> https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/diff/?id=5e25ba5003ee5de0ba2be56bfd54d16d4b1b028d
> 
> I don't know why this TSO patch affect my network since the network
> driver doesn't enable NETIF_F_TSO.
> But I
> (dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
> dev->features |= dev->hw_features;)
> 
> After look into the commit difference,
> I can just doubt if function "tcp_update_pacing_rate" can work well
> with NO_HZ_IDEL or HIGH_RES_TIMERS, since it base on jiffie to
> calculate the rate instead of human nenosecond of new timer
> arch(clocksource/clockevent)
> --

I cant see any effect for this patch explaining a 50% reduction in
speed, unless a special qdisc setup maybe.



--
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