linux-kernel - Performance of "kirqd" in SMP Internet Server

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

Message-ID: <001f01c914a2$4c856d80$e5904880$@snu.ac.kr>
Date:	Fri, 12 Sep 2008 15:39:20 +0900
From:	"Eunsung Kim" <eskim@...lab.snu.ac.kr>
To:	<linux-kernel@...r.kernel.org>
Subject: Performance of "kirqd" in SMP Internet Server 

Hi there.

I am looking for improving network interrupt balancing mechanism in SMP
internet server.
So I have analyzed kirqd in Linux kernel and I have done the following
experiment to realize impact of kirqd.

-------------------------------------------------------------

- Server spec.
CPU: AMD Phenom 9550 Quad-Core (each core 2.2 GHz)
Network card: RealTek Gigabit Ethernet
Network driver: R8169 disabled NAPI

- Client spec.
CPU: Intel Pentium 4 CPU 3.20GHz
Network card: Marvell Gigabit Ethernet

- Network
Connected with gigabit switch

- Benchmark tool
Netperf (throughput for TCP_STREAM in 10 mins)

- Comparison (throughput)
1. kernel 2.6.19.2 (smp_affinity of network irq is CPU 0.)
2. kernel 2.6.19.2 enabled kirqd

-------------------------------------------------------------

Experimental result present that there is no difference between 1 and 2.
Both have achieved throughput of about 940 Mbps via the experiment.
Throughput of 1 describes that netperf traffic can't saturate the server
system and then it is normal traffic.

I have thought that kirqd will 
- get "good" performance due to removing "receive livelock" of a CPU under
high traffic.
- get "poor" performance due to cache miss under low or normal traffic.

But the result for normal traffic doesn't so.
Is it correct?

I had no idea of how to generate high traffic and couldn't perform it.
How can I generate high traffic to saturate the server system?

I wonder whether my experimental setup is correct or not.
How do I have to configure experiment to discover a strong or weak point of
kirqd?

Please comment on my experiment result and setup, and
tell me the web link if there are experiment materials for the system
performance of kirqd

regards.

-- Eunsung

+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=
 Eunsung Kim
 PhD candidate
 Distibuted Computing System Laboratory
 School of Computer Science and Engineering
 Seoul National University
 Tel: +82-2-876-2159
 Email: eskim@...lab.snu.ac.kr
+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+=+= 

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/