Path: archiver1.google.com!news1.google.com!newsfeed.stanford.edu!news.tele.dk!news.tele.dk!small.news.tele.dk!luth.se!uio.no!nntp.uio.no!ifi.uio.no!internet-mailinglist
Newsgroups: fa.linux.kernel
Return-Path: <linux-kernel-owner+fa.linux.kernel=40ifi.uio...@vger.kernel.org>
Original-Date: 	Wed, 26 Feb 2003 15:19:37 -0800
From: "Martin J. Bligh" <mbl...@aracnet.com>
To: linux-kernel <linux-ker...@vger.kernel.org>
cc: lse-tech <lse-t...@lists.sourceforge.net>
Subject: 2.5.63-mjb1 (scalability / NUMA patchset)
Original-Message-ID: <5880000.1046301577@[10.10.2.4]>
X-Mailer: Mulberry/2.2.1 (Linux/x86)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
Sender: linux-kernel-ow...@vger.kernel.org
Precedence: bulk
X-Mailing-List: 	linux-kernel@vger.kernel.org
Organization: Internet mailing list
Date: Wed, 26 Feb 2003 23:20:52 GMT
Message-ID: <fa.crj1pne.22s5gm@ifi.uio.no>
Lines: 176

The patchset contains mainly scalability and NUMA stuff, and anything 
else that stops things from irritating me. It's meant to be pretty stable, 
not so much a testing ground for new stuff.

I'd be very interested in feedback from anyone willing to test on any 
platform, however large or small.

ftp://ftp.kernel.org/pub/linux/kernel/people/mbligh/2.5.63/patch-2.5.63-mjb
1.bz2

additional:

http://www.aracnet.com/~fletch/linux/2.5.59/pidmaps_nodepages

Since 2.5.62-mjb3 (~ = changed, + = added, - = dropped)

Notes:

Merged with Linus:

- fix_was_sched					Ingo / wli / Rick Lindsley
- kirq_clustered_fix				Dave Hansen / Martin J. Bligh

Dropped:

- acpi_x440_hack (proper fix merged)		Anonymous Coward
- auto_disable_tsc				John Stultz

New:

+ acpi_16way					John Stultz
+ schedstat2					Rick Lindsley
+ nfs_fix					Trond Myklebust
+ nonzero_apicid				Martin J. Bligh
+ objrmap_fix					Dave McCracken

Pending:
scheduler callers profiling (Anton)
PPC64 NUMA patches (Anton)
Child runs first (akpm)
Kexec
e1000 fixes
Non-PAE aligned kernel splits (Dave Hansen)
Update the lost timer ticks code
Ingo scheduler updates

Present in this patch:

early_printk					Dave Hansen et al.
	Allow printk before console_init

confighz					Andrew Morton / Dave Hansen
	Make HZ a config option of 100 Hz or 1000 Hz

config_page_offset				Dave Hansen / Andrea
	Make PAGE_OFFSET a config option

vmalloc_stats					Dave Hansen
	Expose useful vmalloc statistics

local_pgdat					William Lee Irwin
	Move the pgdat structure into the remapped space with lmem_map

numameminfo					Martin Bligh / Keith Mannthey
	Expose NUMA meminfo information under /proc/meminfo.numa

notsc						Martin Bligh
	Enable notsc option for NUMA-Q (new version for new config system)

mpc_apic_id					Martin J. Bligh
	Fix null ptr dereference (optimised away, but ...)

doaction					Martin J. Bligh
	Fix cruel torture of macros and small furry animals in io_apic.c

kgdb						Andrew Morton / Various People
	The older version of kgdb, synched with 2.5.54-mm1

noframeptr					Martin Bligh
	Disable -fomit_frame_pointer

ingosched					Ingo Molnar
	Modify NUMA scheduler to have independant tick basis.

schedstat					Rick Lindsley
	Provide stats about the scheduler under /proc/stat

schedstat2					Rick Lindsley
	Provide more stats about the scheduler under /proc/stat

sched_tunables					Robert Love
	Provide tunable parameters for the scheduler (+ NUMA scheduler)

early_ioremap					Dave Hansen
	Provide ioremap in very early boot when we only have 8Mb address space

x440disco_A0					Pat Gaughen / IBM NUMA team
	SLIT/SRAT parsing for x440 discontigmem

acpi_16way					John Stultz
	Make ACPI cope with multi-page something-or-others.

numa_pci_fix					Dave Hansen
	Fix a potential error in the numa pci code from Stanford Checker

pfn_to_nid					William Lee Irwin
	Turn pfn_to_nid into a macro

kprobes						Vamsi Krishna S
	Add kernel probes hooks to the kernel

dmc_exit1					Dave McCracken
	Speed up the exit path, pt 1.

dmc_exit2					Dave McCracken
	Speed up the exit path, pt 1.

shpte						Dave McCracken
	Shared pagetables (as a config option)

thread_info_cleanup (4K stacks pt 1)		Dave Hansen / Ben LaHaise
	Prep work to reduce kernel stacks to 4K
	
interrupt_stacks    (4K stacks pt 2)		Dave Hansen / Ben LaHaise
	Create a per-cpu interrupt stack.

stack_usage_check   (4K stacks pt 3)		Dave Hansen / Ben LaHaise
	Check for kernel stack overflows.

4k_stack            (4K stacks pt 4)		Dave Hansen
	Config option to reduce kernel stacks to 4K

fix_kgdb					Dave Hansen
	Fix interaction between kgdb and 4K stacks

stacks_from_slab				William Lee Irwin
	Take kernel stacks from the slab cache, not page allocation.

thread_under_page				William Lee Irwin
	Fix THREAD_SIZE < PAGE_SIZE case

lkcd						LKCD team
	Linux kernel crash dump support

percpu_loadavg					Martin J. Bligh
	Provide per-cpu loadaverages, and real load averages

irq_affinity					Martin J. Bligh
	Workaround for irq_affinity on clustered apic mode systems (eg x440)

no_kirq						Martin J. Bligh
	Allow disabling of kirq to work properly

cleaner_inodes					Andrew Morton
	Make noatime filesystems more efficient

nfs_fix						Trond Myklebust
	Fix some bug or other in NFS that seems to bite people as a race.

nonzero_apicid					Martin J. Bligh
	Cope with boot cpu != mpstable boot cpu.

partial_objrmap					Dave McCracken
	Object based rmap for filebacked pages.

objrmap_fix					Dave McCracken
	Fix detection of anon pages

-mjb						Martin J. Bligh
	Add a tag to the makefile

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

Path: archiver1.google.com!news1.google.com!newsfeed.stanford.edu!news.tele.dk!news.tele.dk!small.news.tele.dk!newsfeed1.e.nsc.no!nsc.no!nextra.com!uio.no!nntp.uio.no!ifi.uio.no!internet-mailinglist
Newsgroups: fa.linux.kernel
Return-Path: <linux-kernel-owner+fa.linux.kernel=40ifi.uio...@vger.kernel.org>
Original-Date: 	Fri, 28 Feb 2003 10:51:35 -0800
From: "Martin J. Bligh" <mbl...@aracnet.com>
To: linux-kernel <linux-ker...@vger.kernel.org>
cc: lse-tech <lse-t...@lists.sourceforge.net>
Subject: 2.5.63-mjb2 (scalability / NUMA patchset)
Original-Message-ID: <34060000.1046458295@[10.10.2.4]>
In-Reply-To: <5880000.1046301577@[10.10.2.4]>
Original-References:  <5880000.1046301577@[10.10.2.4]>
X-Mailer: Mulberry/2.2.1 (Linux/x86)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
Sender: linux-kernel-ow...@vger.kernel.org
Precedence: bulk
X-Mailing-List: 	linux-kernel@vger.kernel.org
Organization: Internet mailing list
Date: Fri, 28 Feb 2003 18:53:09 GMT
Message-ID: <fa.cgjiav9.h0o88h@ifi.uio.no>
References: <fa.crj1pne.22s5gm@ifi.uio.no>
Lines: 194

The patchset contains mainly scalability and NUMA stuff, and anything 
else that stops things from irritating me. It's meant to be pretty stable, 
not so much a testing ground for new stuff.

I'd be very interested in feedback from anyone willing to test on any 
platform, however large or small.

ftp://ftp.kernel.org/pub/linux/kernel/people/mbligh/2.5.63/patch-2.5.63-mjb2.bz2

additional:

http://www.aracnet.com/~fletch/linux/2.5.59/pidmaps_nodepages

Since 2.5.62-mjb3 (~ = changed, + = added, - = dropped)

Notes:
	Contains some critical objrmap fixes.

Dropped temporarily:
- dmc_exit					Dave McCracken
- shpte						Dave McCracken

New:

+ schedstat-scripts				Rick Lindsley
+ objrmap_fixes					Dave McCracken / Hugh Dickins
+ common_physmap				Andy Whitcroft
+ pfn_to_nid_inline				Andy Whitcroft
+ pfn_valid					Andy Whitcroft
+ numa_x86_pc					Andy Whitcroft
+ physnode_map_u8				Andy Whitcroft
+ profiling_docs				Martin J. Bligh

Pending:
scheduler callers profiling (Anton)
PPC64 NUMA patches (Anton)
Child runs first (akpm)
Kexec
e1000 fixes
Non-PAE aligned kernel splits (Dave Hansen)
Update the lost timer ticks code
Ingo scheduler updates

Present in this patch:

early_printk					Dave Hansen et al.
	Allow printk before console_init

confighz					Andrew Morton / Dave Hansen
	Make HZ a config option of 100 Hz or 1000 Hz

config_page_offset				Dave Hansen / Andrea
	Make PAGE_OFFSET a config option

vmalloc_stats					Dave Hansen
	Expose useful vmalloc statistics

local_pgdat					William Lee Irwin
	Move the pgdat structure into the remapped space with lmem_map

numameminfo					Martin Bligh / Keith Mannthey
	Expose NUMA meminfo information under /proc/meminfo.numa

notsc						Martin Bligh
	Enable notsc option for NUMA-Q (new version for new config system)

mpc_apic_id					Martin J. Bligh
	Fix null ptr dereference (optimised away, but ...)

doaction					Martin J. Bligh
	Fix cruel torture of macros and small furry animals in io_apic.c

kgdb						Andrew Morton / Various People
	The older version of kgdb, synched with 2.5.54-mm1

noframeptr					Martin Bligh
	Disable -fomit_frame_pointer

ingosched					Ingo Molnar
	Modify NUMA scheduler to have independant tick basis.

schedstat					Rick Lindsley
	Provide stats about the scheduler under /proc/schedstat

schedstat2					Rick Lindsley
	Provide more stats about the scheduler under /proc/schedstat

schedstat-scripts				Rick Lindsley
	Provide some scripts for schedstat analysis under scripts/

sched_tunables					Robert Love
	Provide tunable parameters for the scheduler (+ NUMA scheduler)

early_ioremap					Dave Hansen
	Provide ioremap in very early boot when we only have 8Mb address space

x440disco_A0					Pat Gaughen / IBM NUMA team
	SLIT/SRAT parsing for x440 discontigmem

acpi_16way					John Stultz
	Make ACPI cope with multi-page something-or-others.

numa_pci_fix					Dave Hansen
	Fix a potential error in the numa pci code from Stanford Checker

pfn_to_nid					William Lee Irwin
	Turn pfn_to_nid into a macro

kprobes						Vamsi Krishna S
	Add kernel probes hooks to the kernel

# dmc_exit					Dave McCracken
	Speed up the exit path.

# shpte						Dave McCracken
	Shared pagetables (as a config option)

thread_info_cleanup (4K stacks pt 1)		Dave Hansen / Ben LaHaise
	Prep work to reduce kernel stacks to 4K
	
interrupt_stacks    (4K stacks pt 2)		Dave Hansen / Ben LaHaise
	Create a per-cpu interrupt stack.

stack_usage_check   (4K stacks pt 3)		Dave Hansen / Ben LaHaise
	Check for kernel stack overflows.

4k_stack            (4K stacks pt 4)		Dave Hansen
	Config option to reduce kernel stacks to 4K

fix_kgdb					Dave Hansen
	Fix interaction between kgdb and 4K stacks

stacks_from_slab				William Lee Irwin
	Take kernel stacks from the slab cache, not page allocation.

thread_under_page				William Lee Irwin
	Fix THREAD_SIZE < PAGE_SIZE case

lkcd						LKCD team
	Linux kernel crash dump support

percpu_loadavg					Martin J. Bligh
	Provide per-cpu loadaverages, and real load averages

irq_affinity					Martin J. Bligh
	Workaround for irq_affinity on clustered apic mode systems (eg x440)

no_kirq						Martin J. Bligh
	Allow disabling of kirq to work properly

cleaner_inodes					Andrew Morton
	Make noatime filesystems more efficient

nfs_fix						Trond Myklebust
	Fix some bug or other in NFS that seems to bite people as a race.

nonzero_apicid					Martin J. Bligh
	Cope with boot cpu != mpstable boot cpu.

partial_objrmap					Dave McCracken
	Object based rmap for filebacked pages.

objrmap_fix					Dave McCracken
	Fix detection of anon pages

objrmap_fixes					Dave McCracken / Hugh Dickins
	Fix up some mapped sizing bugs in objrmap

common_physmap					Andy Whitcroft
	merge physnode_map implementations from numaq and summit

pfn_to_nid_inline				Andy Whitcroft
	converts the pfn_to_nid macro into an inline

pfn_valid					Andy Whitcroft
	fixes up a bug in copy_page_range

numa_x86_pc					Andy Whitcroft
	adds basic numa support for flat systems

physnode_map_u8					Andy Whitcroft
	converts physnode_map array to u8 (save cache polution)

profiling_docs					Martin J. Bligh
	Basic profiling docs

-mjb						Martin J. Bligh
	Add a tag to the makefile

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

Path: archiver1.google.com!news1.google.com!newsfeed.stanford.edu!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!nntp.infostrada.it!bofh.it!robomod
From: rwh...@earthlink.net
Newsgroups: linux.kernel
Subject: 2.5.63-mjb2 (scalability / NUMA patchset)
Date: Mon, 03 Mar 2003 03:20:09 +0100
Message-ID: <20030303022009$320a@gated-at.bofh.it>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.4i
Sender: robo...@news.nic.it
X-Mailing-List: linux-kernel@vger.kernel.org
Approved: robo...@news.nic.it
Lines: 129
Organization: linux.* mail to news gateway
X-Original-Cc: mbl...@aracnet.com
X-Original-Date: Sun, 2 Mar 2003 21:15:46 -0500
X-Original-Message-ID: <20030303021546.GA4322@rushmore>
X-Original-Sender: linux-kernel-ow...@vger.kernel.org
X-Original-To: linux-ker...@vger.kernel.org

> I'd be very interested in feedback from anyone willing to test on any
> platform, however large or small.

The objrmap patch in recent mjb and mm makes the autoconf-2.52 build (fork)
a few percent faster on uniprocessor K6/2 475 Mhz with 384 MB ram:

1205 seconds 2.5.63-mjb1
1216 seconds 2.5.63-mjb1
1219 seconds 2.5.63-mjb2
1226 seconds 2.5.63-mjb1
1238 seconds 2.5.63-mjb2
1266 seconds 2.5.63-mm1-dline
1267 seconds 2.5.63-mm1-dline
1279 seconds 2.5.63-mm1-dline
1297 seconds 2.5.63
1301 seconds 2.5.63
1309 seconds 2.5.63


The mjb patchset has a HZ=100 option that gives about 1.1% to several tests.
Note the 2.4 kernel and 2.5 with HZ=100 give very similar result:

 kernel                   Tower of Hanoi (unixbench)
 2.5.63-mjb1                     14156.4 (higher is better)
 2.5.63-mjb2                     14156.4
 2.4.21-pre3aa1                  14153.1
 2.5.63                          14002.1
 2.5.63-mm1-dline                14000.2
 2.5.62-mm3                      13999.9

Almost all of the unixbench metrics are a little better with mjb2:

kernel          C Compiler Throughput   Execl Throughput   Copy 1024 bufsize
2.5.63-mjb2                     190.0              368.1              6348.0
2.5.63                          180.7              366.4              6108.0

kernel          Copy 4096 bufsize 	Read 1024 bufsize
2.5.63-mjb2               6391.0             15636.0
2.5.63                    6308.0             14872.0

kernel          Read 4096 bufsize	Write 512 bufsize
2.5.63-mjb2               15775.0             20833.0
2.5.63                    14676.0             20833.0

kernel          Piped Context Switching   Shell Scripts (1)   Shell Scripts (16)   Shell Scripts (64)
2.5.63-mjb2                     73785.0               506.3                 34.0                  8.0
2.5.63                          77893.2               483.4                 32.7                  7.7

kernel          System Call Overhead 
2.5.63-mjb2                 232773.2
2.5.63                      231530.5


LMbench memory latency is a little over 1% lower on kernels with HZ=100:

 Memory latencies in nanoseconds - smaller is better
 ---------------------------------------------------
 kernel                          Mhz     L1 $     L2 $    Main mem
 -----------------------------  -----  -------  -------  ---------
 2.5.63-mjb1                      476     4.20   192.47      262.2
 2.4.21-pre3-ac4                  476     4.20   193.96      262.2
 2.4.21-pre3aa1                   476     4.20   197.31      261.9
 2.5.63-mjb2                      476     4.20   199.25      262.2
 2.5.63-mm1-dline                 476     4.25   195.99      266.5
 2.5.63                           476     4.25   200.82      266.9
 2.5.62-mm3                       476     4.25   204.81      266.7

Several other lmbench tests have approximately 1% improvement.

 AIM7 compute workload was about 1.9% faster with mjb2 compared
to 2.5.63:

 kernel                   Tasks   Jobs/Min         Real    CPU   
 2.4.21-pre3-ac4          384    325.4            7151.7  7129.3
 2.4.21-pre3aa1           384    323.8            7187.6  7160.3
 2.5.63-mjb2              384    322.8            7207.9  7180.6
 2.5.63                   384    319.4            7286.1  7255.7
 2.5.63-mjb1              384    318.6            7303.7  7169.7
 2.5.63-mm1-dline         384    318.0            7318.9  7289.4

On the AIM7 database test, mjb2 looks good compared to the other
2.5.63 kernels.  Recent 2.4 kernels have an edge here as the 
number of tasks goes up, though there is no measurement of 
fairness or interactive usability in this test.

 AIM7 dbase workload
 kernel                   Tasks   Jobs/Min          Real    CPU 
 2.4.21-pre3aa1           4      43.1              551.7   117.1
 2.4.21-pre3-ac4          4      43.1              551.8   116.5
 2.4.21-pre3              4      43.0              552.1   114.6
 2.5.63-mjb2              4      41.8              567.8   124.2
 2.5.63-mm1-dline         4      41.4              574.4   124.8
 2.5.63-mjb1              4      41.2              576.8   122.3
 2.5.63                   4      39.8              597.1   124.4
                                                  
 2.4.21-pre3              8      74.7              636.6   203.1
 2.4.21-pre3-ac4          8      74.3              639.4   203.5
 2.4.21-pre3aa1           8      73.5              646.1   201.5
 2.5.63-mjb2              8      67.9              699.8   212.6
 2.5.63-mjb1              8      67.5              704.3   209.4
 2.5.63-mm1-dline         8      67.0              709.1   208.6
 2.5.63                   8      67.0              709.4   212.0
                                                  
 2.4.21-pre3aa1           12     100.3             710.8   287.1
 2.4.21-pre3              12     97.8              728.7   289.0
 2.4.21-pre3-ac4          12     97.4              731.6   286.8
 2.5.63-mjb2              12     85.0              838.1   302.2
 2.5.63-mm1-dline         12     84.5              843.5   298.1
 2.5.63-mjb1              12     83.8              850.6   292.2
 2.5.63                   12     81.6              873.3   296.0
                                                  
 2.4.21-pre3aa1           16     118.5             801.8   367.4
 2.4.21-pre3-ac4          16     115.9             819.7   371.1
 2.4.21-pre3              16     115.8             821.0   377.1
 2.5.63-mjb2              16     97.7              972.6   386.6
 2.5.63-mm1-dline         16     97.7              972.9   383.0
 2.5.63-mjb1              16     96.7              983.2   376.1
 2.5.63                   16     96.1              989.1   381.3


--
Randy Hron
http://home.earthlink.net/~rwhron/kernel/bigbox.html

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

Path: archiver1.google.com!news1.google.com!newsfeed.stanford.edu!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!newsfeed.icl.net!newsfeed.fjserv.net!news.mailgate.org!nntp.infostrada.it!bofh.it!robomod
From: rwh...@earthlink.net
Newsgroups: linux.kernel
Subject: Re: 2.5.63-mjb2 (scalability / NUMA patchset)
Date: Mon, 03 Mar 2003 14:20:11 +0100
Message-ID: <20030303132011$3d54@gated-at.bofh.it>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
User-Agent: Mutt/1.4i
Sender: robo...@news.nic.it
X-Mailing-List: linux-kernel@vger.kernel.org
Approved: robo...@news.nic.it
Lines: 58
Organization: linux.* mail to news gateway
X-Original-Cc: linux-ker...@vger.kernel.org
X-Original-Date: Mon, 3 Mar 2003 08:19:55 -0500
X-Original-Message-ID: <20030303131955.GA4655@rushmore>
X-Original-Sender: linux-kernel-ow...@vger.kernel.org
X-Original-To: a...@digeo.com

> Pleeeeeeze remember to specify basic things such as the machine,
> amount of memory and especially the filesystem type in use.

>> on uniprocessor K6/2 475 Mhz with 384 MB ram
and two IDE drives on ext2.

Could it be a disk driver issue?  Maybe 2.4 has some IDE
enhancements that aren't in 2.5 yet.  Sounds crazy, but
this is my reasoning... 

On a quad Xeon with QLA2200, the QLogic driver makes the 
biggest difference for AIM7 dbase test.

Here is AIM7 dbase on quad Xeon with 3.75 GB ram over ext2:

AIM7 dbase workload
kernel                   Tasks   Jobs/Min         Real    CPU  
2.5.62-mm2               32     555.9            342.0   155.2 
2.4.21-pre4aa1           32     554.4            342.8   142.3 
2.4.21-pre4aa3           32     551.9            344.4   149.6 
2.4.21-pre4-ac3          32     473.8            401.2   147.7 
2.5.62                   32     473.6            401.3   148.2 
2.5.63-mjb2              32     472.5            402.3   161.5 
2.5.63                   32     471.6            403.1   153.1 
2.2.24-rc3               32     431.9            440.1   165.7 

2.5.62-mm2 has the feral driver.  aa has the QLogic 6.x driver.
Those two kernels rule AIM7 dbase and fserver on quad Xeon with
QLA2200.  I tested earlier 2.5 and aa with/without the newer
QLogic drivers.  It was _the_most_important_ factor for AIM7
dbase and fserver.  Perhaps AIM7 dbase and fserver really suck.
They seem rather impervious to other improvements in the kernel.

> Care to share your aim7 database methodology with me?

AIM7 dbase takes a mixture of AIM9 micro activities and runs 
them in proportion to what it's developers found a circa 1996 
database running.  

My methodology was:

0) Scratch head and look for benchmark that does more than one thing.
1) Stumble on AIM7 which purports to test database, fileserver, compute
   server and multiuser scalability.
2) Run AIM7 dbase to crossover
3) Find the lowest load where the Jobs/Minute is near it's peak.
4) Use result of 3) for max.  Divide into 8 smaller loads.
5) Test all kernels with the same load.

-- 
Randy Hron
http://home.earthlink.net/~rwhron/kernel/bigbox.html

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

Path: archiver1.google.com!news1.google.com!newsfeed.stanford.edu!news-spur1.maxwell.syr.edu!news.maxwell.syr.edu!fu-berlin.de!news.cid.net!bofh.it!robomod
From: Andrew Morton <a...@digeo.com>
Newsgroups: linux.kernel
Subject: Re: 2.5.63-mjb2 (scalability / NUMA patchset)
Date: Mon, 03 Mar 2003 20:20:20 +0100
Message-ID: <20030303192020$7e8b@gated-at.bofh.it>
References: <20030303192020$67da@gated-at.bofh.it>
X-Mailer: Sylpheed version 0.8.9 (GTK+ 1.2.10; i586-pc-linux-gnu)
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
X-Originalarrivaltime: 03 Mar 2003 19:12:17.0936 (UTC) FILETIME=[CEC9D100:01C2E1B8]
Sender: robo...@news.nic.it
X-Mailing-List: linux-kernel@vger.kernel.org
Approved: robo...@news.nic.it
Lines: 61
Organization: linux.* mail to news gateway
X-Original-Cc: linux-ker...@vger.kernel.org
X-Original-Date: Mon, 3 Mar 2003 11:08:34 -0800
X-Original-Message-ID: <20030303110834.031e6d98.akpm@digeo.com>
X-Original-References: <20030303131955.GA4655@rushmore>
X-Original-Sender: linux-kernel-ow...@vger.kernel.org
X-Original-To: rwh...@earthlink.net

rwh...@earthlink.net wrote:
>
> > Pleeeeeeze remember to specify basic things such as the machine,
> > amount of memory and especially the filesystem type in use.
> 
> >> on uniprocessor K6/2 475 Mhz with 384 MB ram
> and two IDE drives on ext2.

Ah, thanks.

> Could it be a disk driver issue?  Maybe 2.4 has some IDE
> enhancements that aren't in 2.5 yet.

Well I tested the AIM7 dbase workload yesterday on 256MB IDE.  2.4 and 2.5
have the same throughput, down to a fraction of one percent.  The entire
working set appeared to be around 200MB so there was no reading from disk at
all.  Just 25 minutes of trickling out very slow O_SYNC writes.  The thing is
dominated by disk seek time.

> Here is AIM7 dbase on quad Xeon with 3.75 GB ram over ext2:
> 
> AIM7 dbase workload
> kernel                   Tasks   Jobs/Min         Real    CPU  
> 2.5.62-mm2               32     555.9            342.0   155.2 
> 2.4.21-pre4aa1           32     554.4            342.8   142.3 
> 2.4.21-pre4aa3           32     551.9            344.4   149.6 
> 2.4.21-pre4-ac3          32     473.8            401.2   147.7 
> 2.5.62                   32     473.6            401.3   148.2 
> 2.5.63-mjb2              32     472.5            402.3   161.5 
> 2.5.63                   32     471.6            403.1   153.1 
> 2.2.24-rc3               32     431.9            440.1   165.7 
> 
> 2.5.62-mm2 has the feral driver.  aa has the QLogic 6.x driver.

Well if there is any difference in drive caching policy then one would expect
to see large differences.  Using writeback caching in the disk (which is
considered cheating) would speed things up.  But I'd be surprised if
2.5-vs-2.4 IDE affected the drive's caching policy.

> Those two kernels rule AIM7 dbase and fserver on quad Xeon with
> QLA2200.  I tested earlier 2.5 and aa with/without the newer
> QLogic drivers.  It was _the_most_important_ factor for AIM7
> dbase and fserver.  Perhaps AIM7 dbase and fserver really suck.
> They seem rather impervious to other improvements in the kernel.

Yes, they do.

> > Care to share your aim7 database methodology with me?
> 
> AIM7 dbase takes a mixture of AIM9 micro activities and runs 
> them in proportion to what it's developers found a circa 1996 
> database running.  

AIM7 dbase would probably be more interesting if it created a larger working
set.

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

			  SCO's Case Against IBM

November 12, 2003 - Jed Boal from Eyewitness News KSL 5 TV provides an
overview on SCO's case against IBM. Darl McBride, SCO's president and CEO,
talks about the lawsuit's impact and attacks. Jason Holt, student and 
Linux user, talks about the benefits of code availability and the merits 
of the SCO vs IBM lawsuit. See SCO vs IBM.

Note: The materials and information included in these Web pages are not to
be used for any other purpose other than private study, research, review
or criticism.