Date: Wed, 28 Jan 2004 12:50:43 -0600 From: Jeremy Messenger <mezz7@cox.net> To: John Baldwin <jhb@FreeBSD.org> Cc: Peter Losher <Peter_Losher@isc.org> Subject: Re: spin lock panic in 5.2-REL Message-ID: <opr2h1atax8ckrg5@smtp.central.cox.net> In-Reply-To: <200401281243.10655.jhb@FreeBSD.org> References: <20040127082106.L14045@farside.isc.org> <200401271334.07761.jhb@FreeBSD.org> <200401271420.05240.Peter_Losher@isc.org> <200401281243.10655.jhb@FreeBSD.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, 28 Jan 2004 12:43:10 -0500, John Baldwin <jhb@FreeBSD.org> wrote:
> On Tuesday 27 January 2004 05:19 pm, Peter Losher wrote:
>> On Tuesday 27 January 2004 10:34 am, John Baldwin wrote:
>> > This is a very hard to debug problem. If WITNESS is on it provides
>> more
>> > useful information that can help track that down, but I wouldn't
>> expect a
>> > production machine to be using WITNESS.
>>
>> Yes, I was running a kernel for a short while w/ INVARIANTS and WITNESS,
>> but it was basically useless after a couple of hours under load under
>> the
>> giant lock. How much of a performance hit is just WITNESS?
>
> A lot. I think Robert Watson has seen as high a performance impact as
> 40%.
I have played with the bytebench few days ago, just for fun... This
machine that I played is 464Mhz and 386mb ram; it uses libkse and ULE
(four days ago -CURRENT).
W/OUT-DEBUG = Remove all debug options from kernel and set
/etc/malloc.conf->aj..
=================================
--W/-DEBUG-- -W/OUT-DEBUG-
Dhrystone 2 without register variables 1105081.4 lps 1097325.0 lps
Dhrystone 2 using register variables 1100857.8 lps 1109398.3 lps
Arithmetic Test (type = arithoh) 2076290.3 lps 2086816.9 lps
Arithmetic Test (type = register) 125288.4 lps 126051.2 lps
Arithmetic Test (type = short) 125818.8 lps 126406.5 lps
Arithmetic Test (type = int) 125481.3 lps 126101.5 lps
Arithmetic Test (type = long) 125293.6 lps 126067.5 lps
Arithmetic Test (type = float) 145203.2 lps 146956.6 lps
Arithmetic Test (type = double) 145017.1 lps 145718.5 lps
System Call Overhead Test 11830.6 lps 109590.1 lps
Pipe Throughput Test 20572.7 lps 96916.4 lps
Pipe-based Context Switching Test 6856.2 lps 25128.9 lps
Process Creation Test 155.1 lps 865.2 lps
Execl Throughput Test 62.7 lps 261.0 lps
File Read (10 seconds) 40538.0 KBps 246172.0 KBps
File Write (10 seconds) 7033.0 KBps 11565.0 KBps
File Copy (10 seconds) 4907.0 KBps 12555.0 KBps
File Read (30 seconds) 40688.0 KBps 246785.0 KBps
File Write (30 seconds) 7022.0 KBps 11343.0 KBps
File Copy (30 seconds) 4763.0 KBps 12400.0 KBps
C Compiler Test 187.1 lpm 327.1 lpm
Shell scripts (1 concurrent) 98.0 lpm 623.6 lpm
Shell scripts (2 concurrent) 50.6 lpm 326.3 lpm
Shell scripts (4 concurrent) 26.0 lpm 168.0 lpm
Shell scripts (8 concurrent) 13.0 lpm 82.0 lpm
Dc: sqrt(2) to 99 decimal places 5299.0 lpm 31721.7 lpm
Recursion Test--Tower of Hanoi 11543.3 lps 11618.8 lps
INDEX VALUES ---------W/-DEBUG---------
--------W/OUT-DEBUG-------
TEST BASELINE RESULT INDEX
BASELINE RESULT INDEX
Arithmetic Test (type = double) 2541.7 145017.1 57.1
2541.7 145718.5 57.3
Dhrystone 2 without register variables 22366.3 1105081.4 49.4
22366.3 1097325.0 49.1
Execl Throughput Test 16.5 62.7 3.8
16.5 261.0 15.8
File Copy (30 seconds) 179.0 4763.0 26.6
179.0 12400.0 69.3
Pipe-based Context Switching Test 1318.5 6856.2 5.2
1318.5 25128.9 19.1
Shell scripts (8 concurrent) 4.0 13.0 3.2
4.0 82.0 20.5
=========
=========
SUM of 6 items
145.3 231.0
AVERAGE
24.2 38.5
=================================
Cheers,
Mezz
--
bsdforums.org 's moderator, mezz.
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?opr2h1atax8ckrg5>