netdev - Re: query: localhost - 794ed393b clips hefty load tbench, does it matter?

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: <1362068018.15793.31.camel@edumazet-glaptop>
Date:	Thu, 28 Feb 2013 08:13:38 -0800
From:	Eric Dumazet <eric.dumazet@...il.com>
To:	Mike Galbraith <bitbucket@...ine.de>
Cc:	netdev <netdev@...r.kernel.org>, Eric Dumazet <edumazet@...gle.com>
Subject: Re: query: localhost - 794ed393b clips hefty load tbench, does it
 matter?

On Thu, 2013-02-28 at 13:49 +0100, Mike Galbraith wrote:
> Greetings network wizards,
> 
> I was testing a 64 core box after 3.0-stable update, and noticed
> $subject.
> 
> vogelweide:~/:[0]# numactl --hardware
> available: 1 nodes (0)
> node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63
> node 0 size: 8181 MB
> node 0 free: 7353 MB
> node distances:
> node   0 
>   0:  10
> 
> Sob, poor thing.  Anyway, that's the box in case it matters.
> 
> Without 94ed393b. 
> 
> vogelweide:~/:[0]# for i in 1 2 4 8 16 32 64 128 256 512; do tbench.sh $i 10 2>&1|grep Throughput; done
> Throughput 288.784 MB/sec 1 procs
> Throughput 559.937 MB/sec 2 procs
> Throughput 1068.75 MB/sec 4 procs
> Throughput 2159.04 MB/sec 8 procs
> Throughput 4193.75 MB/sec 16 procs
> Throughput 7662.24 MB/sec 32 procs
> Throughput 9034.49 MB/sec 64 procs
> Throughput 9045.9 MB/sec 128 procs
> Throughput 9077.55 MB/sec 256 procs
> Throughput 8907.48 MB/sec 512 procs
> 
> With.
> 
> vogelweide:~/:[0]# for i in 1 2 4 8 16 32 64 128 256 512; do tbench.sh $i 10 2>&1|grep Throughput; done
> Throughput 288.833 MB/sec 1 procs
> Throughput 520.87 MB/sec 2 procs
> Throughput 937.758 MB/sec 4 procs
> Throughput 1563.3 MB/sec 8 procs
> Throughput 1775.14 MB/sec 16 procs
> Throughput 1406.55 MB/sec 32 procs
> Throughput 1448.77 MB/sec 64 procs
> Throughput 1468.92 MB/sec 128 procs
> Throughput 1525.35 MB/sec 256 procs
> Throughput 1713.54 MB/sec 512 procs
> 
> I'm wondering if this could cause problems on a big box doing something
> like say mysql queries of a local database, blasting retrieved data out
> over industrial strength copper/glass or such.  My desktop box surely
> won't notice, but it seems heavy lifters might.  I saw the reason for
> it, but I was left wondering why we used to care about it, but no more,
> so here I am to see if I can get my curiosity spot scratched.
> 
> I'll sorta miss good ole tbench in scheduler litmus test role.  On the
> bright side, localhost based scalability reports are history.  Oh wait.

Sure, this patch re-introduces the dst->__refcnt false sharing for
loopback.

Hopefully, with current kernels it's not an issue, because each cpu gets
a different dst.

(It would be an issue if the connect() calls are all done on a single
cpu, than traffic handled on other cpus)

So please try tbench on linux-3.8 or current git tree ;)



--
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