Re: ULE 2.0


On Sun, 7 Jan 2007, Massimo Lusetti wrote:

On Sat, 6 Jan 2007 04:15:13 -0800 (PST)
Jeff Roberson <jroberson@xxxxxxxxxxxxxx> wrote:


I step in just to show some results i got with super-smack.

This is -CURRENT as of yesterday (06/01) with mysql 5.1.12 from
packages and super-smack 1.13 from ports.
I built two kernels, the GENERIC and a GENERIC with ULE instead of 4BSD
which i named GENEULE. I got dmesg from both if you're interested.

Thanks, I am working on some changes to improve SMP load balancing. I'm going to test them more thoroughly on an array of machines that I have here before I commit them, but I'll let you know when I do.

In the future, it's more useful to do performance comparisons without WITNESS or INVARIANTS enabled as these options may unfairly hinder one scheduler or the other. However, I don't doubt the relative results you present.

Thanks,
Jeff


The test box is my laptop which is seens as:

Copyright (c) 1992-2007 The FreeBSD Project.
Copyright (c) 1979, 1980, 1983, 1986, 1988, 1989, 1991, 1992, 1993, 1994
The Regents of the University of California. All rights
reserved. FreeBSD is a registered trademark of The FreeBSD Foundation.
FreeBSD 7.0-CURRENT #0: Sun Jan 7 13:19:24 CET 2007
root@xxxxxxxxxxxxxxxxxxx:/usr/obj/usr/src/sys/GENEULE
WARNING: WITNESS option enabled, expect reduced performance.
ACPI APIC Table: <Acer Grape >
Timecounter "i8254" frequency 1193182 Hz quality 0
CPU: Genuine Intel(R) CPU T2300 @ 1.66GHz (1662.52-MHz
686-class CPU) Origin = "GenuineIntel" Id = 0x6e8 Stepping = 8
Features=0xbfe9fbff<FPU,VME,DE,PSE,TSC,MSR,PAE,MCE,CX8,APIC,SEP,MTRR,PGE,MCA,CMOV,PAT,CLFLUSH,DTS,ACPI,MMX,FXSR,SSE,SSE2,SS,HTT,TM,PBE>
Features2=0xc1a9<SSE3,MON,VMX,EST,TM2,XTPR,<b15>>
AMD Features=0x100000<NX>
Cores per package: 2
real memory = 1072168960 (1022 MB)
avail memory = 1035640832 (987 MB)
FreeBSD/SMP: Multiprocessor System Detected: 2 CPUs
cpu0 (BSP): APIC ID: 0
cpu1 (AP): APIC ID: 1



During the test the machine was completely idle, expect for super-smack
and mysqld of course.

The tests cases where select-key and update-select always
with 10 clients and 10000 queries each executed 5 times.
On every threading library switch the server were restarted.

Here the results in order as they were executed.




Query_type num_queries max_time min_time q_per_s

4BSD libpthread

select_index 200000 2 0 4817.59
select_index 200000 2 0 4793.25
select_index 200000 3 0 4814.93
select_index 200000 2 0 4770.63
select_index 200000 2 0 4819.42
--
update_index 100000 2 0 2131.67
update_index 100000 2 0 2168.52
update_index 100000 3 0 2170.03
update_index 100000 2 0 2169.04
update_index 100000 3 0 2166.57


4BSD libthr

select_index 200000 1 0 5079.52
select_index 200000 3 0 4728.71
select_index 200000 3 0 4707.58
select_index 200000 1 0 4682.04
select_index 200000 1 0 4687.51
--
update_index 100000 1 0 2323.26
update_index 100000 2 0 2315.49
update_index 100000 2 0 2365.37
update_index 100000 2 0 2324.08
update_index 100000 2 0 2323.27


ULE libpthread

select_index 200000 3 0 4487.76
select_index 200000 0 0 4419.15
select_index 200000 3 0 4104.53
select_index 200000 3 0 4083.51
select_index 200000 4 0 4063.64
--
update_index 100000 3 0 1525.89
update_index 100000 3 0 1532.17
update_index 100000 4 0 1531.36
update_index 100000 4 0 1530.10
update_index 100000 4 0 1520.19


ULE libthr
select_index 200000 3 0 4318.65
select_index 200000 8 0 4347.53
select_index 200000 13 0 4327.46
select_index 200000 2 0 4312.92
select_index 200000 3 0 4349.71
--
update_index 100000 3 0 1710.85
update_index 100000 2 0 1731.89
update_index 100000 3 0 1716.59
update_index 100000 3 0 1711.15
update_index 100000 3 0 1697.40



Thanks to all your work and for reading this till the end ;) now i wait
to read you further

Regards
--
Massimo.run();
THE DAILY PLANET SUPERMAN SAVES DESSERT! Plans to "Eat it later"

_______________________________________________
freebsd-current@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-current
To unsubscribe, send any mail to "freebsd-current-unsubscribe@xxxxxxxxxxx"

Relevant Pages

  • Re: ULE 2.0
    ... I have a dual core opteron that works ... It is mysql super-smack, I am trying to get a performance result. ... the GENERIC and a GENERIC with ULE instead of 4BSD ... FreeBSD is a registered trademark of The FreeBSD Foundation. ...
    (freebsd-current)
  • cvs-src summary for September 6-13
    ... It is intended to help the FreeBSD community keep up with the fast-paced ... You can get old summaries, and an HTML version of this one, at ... Scott Long changed the default scheduler from the new ULE one ... scheduler bugs preventing it from being used in the release`. ...
    (freebsd-current)
  • Re: 2 x quad-core system is slower that 2 x dual core on FreeBSD
    ... larger then about 2 and the loads include much kernel work (e.g. IO, ... I know this thread is about SMP scalling, but most of my machines are UP so I run my own synthetic benchmarks (super-smack on MySQL 5.0.45 and ab on Apache 2.2.6) on an old box with AMD Barton 2500+ with 512MB RAM. ... because FreeBSD 6.2 UP behaves better than FreeBSD 7.0-BETA3 (4BSD and ULE tested). ...
    (freebsd-stable)
  • Re: SchedULE vs BSD scheduler
    ... If you install FreeBSD 7, ... What's the advantage of ULE / disadvantage of the default? ... SMP systems (but it seems to improve interactivity on ... Handelsregister: Registergericht Muenchen, HRA 74606, Geschäftsfuehrung: ...
    (freebsd-stable)
  • Re: Bug in calcru in he 6.2 and 6.3 kernels
    ... | Murty, Ravi wrote: ... The problem was that (and I am using ULE) sched_bind ... |> acceptable (i.e. it is okay to be running and oncpu is NOCPU). ... FreeBSD - The Power to Serve! ...
    (freebsd-hackers)