* [PATCHSET] Linux 2.4.18-rc3-jam1
@ 2002-02-22 0:50 J.A. Magallon
2002-02-22 7:38 ` Barry K. Nathan
0 siblings, 1 reply; 11+ messages in thread
From: J.A. Magallon @ 2002-02-22 0:50 UTC (permalink / raw)
To: Lista Linux-Kernel; +Cc: rwhron
Hi.
I have updatd my pathset with latest bits for 2.4.18-rc3.
Main contents:
- vm-25, vm_io-2
- sched-O1-K3, sched-balance-fixes
- lowlatency-mini, read-latency-2
- aic-6.2.5, ide-20020215
Extras:
- ide-cd-dma
- some minor bits-optimizations for SMP from -aa
- irqrate-A1, intr-seq-file
- i2c+sensors 2.6.2 CVS
- bproc-3.1.7
You can get it at
http://giga.cps.unizar.es/~magallon/linux/kernel/2.4.18-rc3-jam1/
Try and enjoy...(at least my box is running fine).
--
J.A. Magallon # Let the source be with you...
mailto:jamagallon@able.es
Mandrake Linux release 8.2 (Cooker) for i586
Linux werewolf 2.4.18-rc3-jam1 #1 SMP Fri Feb 22 01:11:56 CET 2002 i686
^ permalink raw reply [flat|nested] 11+ messages in thread
* [PATCHSET] Linux 2.4.18-rc3-jam1
@ 2002-02-22 4:21 rwhron
2002-02-22 5:35 ` Andrew Morton
0 siblings, 1 reply; 11+ messages in thread
From: rwhron @ 2002-02-22 4:21 UTC (permalink / raw)
To: jamagallon; +Cc: andrea, linux-kernel
> Hi.
>
> I have updatd my pathset with latest bits for 2.4.18-rc3.
I'm looking forward to trying your patchsets more in the
future. I was consolidating some results and it made
sense to include 2.4.18-rc2-jam1.
2.4.18-rc2-jam1 was the only kernel of these three that had Andrew
Morton's read_latency2 patch. 2.4.18-rc2-jam1 was compiled with
vm-24, vm_io-2, sched-O1-K3, sched-aa-fixes, lowlatency-mini,
read-latency2, and sched-balance applied.
Tiobench read/write latency is where the biggest differences are.
K6-2 475 mhz, 384 MB ram, IDE disks.
dbench 128 processes
2.4.18-rc2-jam1 ***************************** 14.9 MB/sec
2.4.18rc2aa2 ***************************** 14.8 MB/sec
2.4.18-rc2 *************** 7.6 MB/sec
Unixbench-4.2.0 2.4.18rc2aa2 2.4.18-rc2 2.4.18-rc2-jam1
Execl Throughput 381.0 303.7 299.0 lps
Pipe Throughput 389525.7 386034.4 324554.7 lps
Pipe-based Context Switching 94894.8 84108.7 82547.7 lps
Process Creation 1398.1 1101.3 1239.3 lps
System Call Overhead 323473.1 335228.1 336939.0 lps
Shell Scripts (1 concurrent) 698.0 612.1 620.7 lpm
Shell Scripts (16 concurrent) 48.0 42.7 42.0 lpm
Shell Scripts (32 concurrent) 24.0 21.3 20.7 lpm
Shell Scripts (64 concurrent) 11.8 10.4 10.0 lpm
LMbench 2.0p2 average of 9 runs
-------------------------------
Processor, Processes - times in microseconds - smaller is better
null open selct sig sig fork exec sh
I/O stat clos TCP inst hndl proc proc proc
2.4.18-rc2 0.72 3.10 4.78 39.0 1.48 3.33 913 3578 13555
2.4.18-rc2-jam1 0.78 3.05 4.46 37.3 1.70 3.09 887 3530 13434
2.4.18rc2aa2 0.76 3.15 4.94 59.4 1.43 3.10 740 2791 11178
Context switching - times in microseconds - smaller is better
2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K
ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw
2.4.18-rc2 2.73 22.57 194.41 56.61 207.56 60.40 225.63
2.4.18-rc2-jam1 1.21 18.68 180.87 51.49 205.42 56.37 220.32
2.4.18rc2aa2 1.42 22.81 184.62 57.22 206.74 59.03 226.33
*Local* Communication latencies in microseconds - smaller is better
Pipe AF UDP RPC/ TCP RPC/ TCP
UNIX UDP TCP conn
2.4.18-rc2 10.87 21.70 41.55 141.2 83.87 190.0 290.36
2.4.18-rc2-jam1 12.93 20.95 42.77 145.4 79.17 203.0 305.82
2.4.18rc2aa2 10.86 20.40 38.09 129.3 77.17 194.2 259.29
File & VM system latencies in microseconds - smaller is better
0K File 10K File Mmap Prot Page
Create Delete Create Delete Latency Fault Fault
2.4.18-rc2 126.85 184.96 639.46 262.70 2630.3 1.10 6.8
2.4.18-rc2-jam1 131.84 181.95 684.64 260.11 2648.1 1.02 7.0
2.4.18rc2aa2 122.44 174.66 611.18 248.13 2659.6 1.29 5.3
*Local* Communication bandwidths in MB/s - bigger is better
Pipe AF TCP File Mmap Bcopy Bcopy Mem Mem
UNIX reread reread (libc) (hand) read write
2.4.18-rc2 70.10 42.07 60.04 61.66 237.60 59.32 60.48 237.48 85.32
2.4.18-rc2-jam1 65.96 45.27 48.45 59.49 237.50 59.06 60.21 237.52 84.75
2.4.18rc2aa2 64.89 36.07 53.51 62.22 237.66 59.17 60.33 237.51 84.83
Memory latencies in nanoseconds - smaller is better
Mhz L1 $ L2 $ Main mem
2.4.18-rc2 476 4.20 192.34 261.99
2.4.18-rc2-jam1 476 4.20 196.10 262.02
2.4.18rc2aa2 476 4.20 191.15 261.97
Tiobench average of 3 runs
--------------------------
2048 MB worth of files on ext2 fs.
Read, write, and seek rates in MB/sec.
Latency in milliseconds.
Percent of requests that took longer than 2 and 10 seconds.
Sequential Reads
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
----------------- --- ------------------------------------------------------
2.4.18-rc2 8 9.08 11.75% 10.271 688.32 0.00000 0.00000 77
2.4.18-rc2-jam1 8 9.32 11.99% 10.029 592.48 0.00000 0.00000 78
2.4.18rc2aa2 8 8.70 11.05% 10.721 717.32 0.00000 0.00000 79
2.4.18-rc2 16 9.15 11.49% 20.295 1181.92 0.00000 0.00000 80
2.4.18-rc2-jam1 16 9.66 12.19% 19.346 935.39 0.00000 0.00000 79
2.4.18rc2aa2 16 7.81 9.50% 23.910 1395.18 0.00000 0.00000 82
2.4.18-rc2 32 9.05 11.50% 38.686 137791.12 0.01105 0.00991 79
2.4.18-rc2-jam1 32 9.54 12.37% 39.121 1720.87 0.00000 0.00000 77
2.4.18rc2aa2 32 7.13 8.69% 41.084 573101.95 0.00630 0.00630 82
2.4.18-rc2 64 8.34 10.65% 66.858 372960.69 0.04826 0.04730 78
2.4.18-rc2-jam1 64 9.85 13.41% 75.739 3364.59 0.00000 0.00000 73
2.4.18rc2aa2 64 6.45 7.61% 77.403 605856.31 0.03337 0.03318 85
2.4.18-rc2 128 8.06 10.25% 117.062 496375.50 0.10891 0.10128 79
2.4.18-rc2-jam1 128 9.77 17.22% 150.658 6822.95 0.83160 0.00000 57
2.4.18rc2aa2 128 6.04 7.29% 144.083 730872.49 0.07763 0.07763 83
Random Reads
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
----------------- --- ------------------------------------------------------
2.4.18-rc2 8 0.48 1.26% 192.062 594.20 0.00000 0.00000 38
2.4.18-rc2-jam1 8 0.48 0.82% 193.114 586.86 0.00000 0.00000 59
2.4.18rc2aa2 8 0.47 1.14% 194.488 640.23 0.00000 0.00000 41
2.4.18-rc2 16 0.53 1.38% 343.392 1021.49 0.00000 0.00000 38
2.4.18-rc2-jam1 16 0.53 1.30% 341.295 1040.82 0.00000 0.00000 41
2.4.18rc2aa2 16 0.51 1.51% 354.999 965.62 0.00000 0.00000 34
2.4.18-rc2 32 0.57 1.21% 599.111 1927.88 0.00000 0.00000 47
2.4.18-rc2-jam1 32 0.57 1.46% 627.710 1887.15 0.00000 0.00000 39
2.4.18rc2aa2 32 0.50 1.60% 642.256 1439.27 0.00000 0.00000 31
2.4.18-rc2 64 0.60 1.92% 1000.255 58973.91 0.20161 0.20161 31
2.4.18-rc2-jam1 64 0.61 2.17% 1149.896 3079.41 0.00000 0.00000 28
2.4.18rc2aa2 64 0.49 1.33% 1197.580 79710.81 0.75606 0.75606 37
2.4.18-rc2 128 0.60 1.77% 1655.593 58576.80 4.91431 4.51109 34
2.4.18-rc2-jam1 128 0.63 3.36% 2177.767 5866.58 0.02520 0.00000 19
2.4.18rc2aa2 128 0.51 1.25% 1749.619 74096.20 5.29233 5.26713 41
Sequential Writes
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
----------------- --- ------------------------------------------------------
2.4.18-rc2 8 13.95 35.53% 6.368 10257.41 0.02461 0.00000 39
2.4.18-rc2-jam1 8 18.55 86.03% 4.501 8531.84 0.00267 0.00000 22
2.4.18rc2aa2 8 17.20 71.51% 4.523 3930.59 0.00000 0.00000 24
2.4.18-rc2 16 13.20 33.72% 13.478 19726.28 0.17261 0.00000 39
2.4.18-rc2-jam1 16 18.45 83.61% 8.865 15315.44 0.30327 0.00000 22
2.4.18rc2aa2 16 15.55 60.43% 10.153 6891.82 0.00172 0.00000 26
2.4.18-rc2 32 13.16 35.35% 25.744 33739.34 0.46578 0.00190 37
2.4.18-rc2-jam1 32 18.29 80.37% 16.850 31111.62 0.30365 0.00381 23
2.4.18rc2aa2 32 12.96 41.80% 24.453 11125.65 0.02250 0.00000 31
2.4.18-rc2 64 13.60 44.32% 46.701 50005.93 0.70420 0.04386 31
2.4.18-rc2-jam1 64 18.09 79.76% 30.954 105671.85 0.30422 0.16518 23
2.4.18rc2aa2 64 11.16 33.81% 56.135 18174.69 0.98952 0.00000 33
2.4.18-rc2 128 12.98 44.74% 86.349 67514.69 1.12170 0.25120 29
2.4.18-rc2-jam1 128 18.02 79.18% 55.833 115163.05 0.28268 0.26455 23
2.4.18rc2aa2 128 10.54 31.31% 113.671 26464.76 2.84768 0.00000 34
Random Writes
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
----------------- --- ------------------------------------------------------
2.4.18-rc2 8 0.56 1.08% 1.267 265.14 0.00000 0.00000 52
2.4.18-rc2-jam1 8 0.71 4.71% 0.190 3.53 0.00000 0.00000 15
2.4.18rc2aa2 8 0.57 1.02% 0.581 1534.89 0.00000 0.00000 56
2.4.18-rc2 16 0.57 1.13% 1.760 430.99 0.00000 0.00000 50
2.4.18-rc2-jam1 16 0.72 4.76% 0.192 4.02 0.00000 0.00000 15
2.4.18rc2aa2 16 0.57 1.10% 4.640 7059.08 0.15000 0.00000 52
2.4.18-rc2 32 0.60 1.18% 1.735 566.45 0.00000 0.00000 51
2.4.18-rc2-jam1 32 0.77 5.13% 0.192 5.28 0.00000 0.00000 15
2.4.18rc2aa2 32 0.58 1.28% 12.224 18987.85 0.22500 0.00000 45
2.4.18-rc2 64 0.65 1.40% 1.477 553.19 0.00000 0.00000 46
2.4.18-rc2-jam1 64 0.79 5.27% 0.190 3.63 0.00000 0.00000 15
2.4.18rc2aa2 64 0.59 1.17% 18.645 25314.00 0.17641 0.10080 51
2.4.18-rc2 128 0.67 1.87% 1.334 777.23 0.00000 0.00000 36
2.4.18-rc2-jam1 128 0.80 5.72% 0.190 3.68 0.00000 0.00000 14
2.4.18rc2aa2 128 0.61 1.39% 61.796 72674.58 0.32761 0.32761 44
bonnie++-1.02a on 1024 MB file
---------------------Sequential Output--------------------
-----Per Char----- ------Block------- -----Rewrite------
MB/sec %CPU Eff MB/sec %CPU Eff MB/sec %CPU Eff
2.4.18-rc2 3.43 98.0 3.50 13.88 57.0 24.35 8.04 44.3 18.13
2.4.18-rc2-jam1 3.16 87.0 3.63 15.27 81.0 18.85 6.35 30.0 21.17
2.4.18rc2aa2 3.41 96.3 3.53 15.30 88.3 17.32 6.62 22.3 29.64
-----------Sequential Input----------- ------Random-----
-----Per Char----- ------Block------- ------Seeks------
MB/sec %CPU Eff MB/sec %CPU Eff /sec %CPU Eff
2.4.18-rc2 3.99 97.3 4.10 15.61 56.0 27.87 129 2.0 6451
2.4.18-rc2-jam1 3.42 82.0 4.17 13.75 17.3 79.35 122 2.0 6116
2.4.18rc2aa2 3.35 80.3 4.17 12.96 23.0 56.35 127 1.7 7596
Bonnie++ on 16384 small files
-------Sequential Create------------
------Create----- -----Delete-----
/sec %CPU Eff /sec %CPU Eff
2.4.18-rc2 4266 98.0 4353 4456 96.7 4609
2.4.18-rc2-jam1 4502 97.7 4609 4592 99.0 4638
2.4.18rc2aa2 4353 99.0 4397 4335 94.7 4579
------------Random Create-------------
------Create----- -----Delete-----
/sec %CPU Eff /sec %CPU Eff
2.4.18-rc2 4216 98.0 4302 3826 91.0 4204
2.4.18-rc2-jam1 4534 99.0 4580 3830 91.7 4177
2.4.18rc2aa2 4392 98.0 4481 4277 99.7 4291
OSDB is the time in seconds to run the single and multi-user (5) open source
database benchmark on postgresql-7.2.
Time to build/run autoconf perl kernel updatedb updatedb5 OSDB
2.4.18rc2aa2 1087 1582 1372 40 43 7961
2.4.18-rc2-jam1 1180 1658 1333 31 46 10064
2.4.18-rc2 1187 1620 1391 30 45 8802
Memory usage available kern_code reserved data
2.4.18rc2aa2 384624k 895k 8208k 212k
2.4.18-rc2 384296k 887k 8536k 209k
2.4.18-rc2-jam1 384288k 891k 8544k 210k
Testing details and more results at:
http://home.earthlink.net/~rwhron/kernel/k6-2-475.html
--
Randy Hron
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
2002-02-22 4:21 rwhron
@ 2002-02-22 5:35 ` Andrew Morton
0 siblings, 0 replies; 11+ messages in thread
From: Andrew Morton @ 2002-02-22 5:35 UTC (permalink / raw)
To: rwhron; +Cc: jamagallon, andrea, linux-kernel
rwhron@earthlink.net wrote:
>
> ...
> Tiobench average of 3 runs
> --------------------------
> ...
> Random Writes
> Num Avg Maximum Lat% Lat% CPU
> Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
> ----------------- --- ------------------------------------------------------
> ...
> 2.4.18-rc2 128 0.67 1.87% 1.334 777.23 0.00000 0.00000 36
> 2.4.18-rc2-jam1 128 0.80 5.72% 0.190 3.68 0.00000 0.00000 14
> 2.4.18rc2aa2 128 0.61 1.39% 61.796 72674.58 0.32761 0.32761 44
>
> ...
Holy cow! Are you sure these numbers are right?
The increased throughput will be thanks to the boosted request
queue size.
The (greatly) increased CPU load will also be due to browsing the eight-times
larger request queue. Plus we browse it a bit more than we used to.
The improvement in worst-case latency in both -aa and -jam will
be due to the FIFO wait for requests.
But improvement by a factor of 20,000 sounds a little excessive :)
And a maximum latency of three milliseconds would seem to indicate
that the benchmark is *never* waiting on disk seek, and that
perhaps the request queue is simply never filling up. But that
doesn't make sense.
What does the "latency" actually mean? Is it the time spent
in the kernel to issue a write(2)?
Something funny is happening, I suspect. Guess I should go
look at tiobench...
-
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
2002-02-22 0:50 [PATCHSET] Linux 2.4.18-rc3-jam1 J.A. Magallon
@ 2002-02-22 7:38 ` Barry K. Nathan
2002-02-23 1:33 ` J.A. Magallon
0 siblings, 1 reply; 11+ messages in thread
From: Barry K. Nathan @ 2002-02-22 7:38 UTC (permalink / raw)
To: J.A. Magallon; +Cc: Lista Linux-Kernel, rwhron
> Extras:
[snip]
> - irqrate-A1, intr-seq-file
FWIW, the irqrate-A1 patch from the -slb/-jam series kernels
(2.4.18-rc1-slb2 through 2.4.18-rc2-jam1 for certain) seems to make one of
my computers (a dual 1GHz PIII, Iwill DVD266-R motherboard) hang when I
access the floppy drive. I don't believe this happens on other
(uniprocessor, not SMP) machines I have here, although I'm not certain
about that. This is 100% reproducible.
This bug report is probably rather vague, but I probably won't have time
for anything more detailed or any debugging activities related to this for
several weeks, and I hope this report is better than nothing.
-Barry K. Nathan <barryn@pobox.com>
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
2002-02-22 7:38 ` Barry K. Nathan
@ 2002-02-23 1:33 ` J.A. Magallon
2002-02-23 8:23 ` Barry K. Nathan
0 siblings, 1 reply; 11+ messages in thread
From: J.A. Magallon @ 2002-02-23 1:33 UTC (permalink / raw)
To: Barry K. Nathan; +Cc: Lista Linux-Kernel, rwhron
On 20020222 Barry K. Nathan wrote:
>> Extras:
>[snip]
>> - irqrate-A1, intr-seq-file
>
>FWIW, the irqrate-A1 patch from the -slb/-jam series kernels
>(2.4.18-rc1-slb2 through 2.4.18-rc2-jam1 for certain) seems to make one of
>my computers (a dual 1GHz PIII, Iwill DVD266-R motherboard) hang when I
>access the floppy drive. I don't believe this happens on other
>(uniprocessor, not SMP) machines I have here, although I'm not certain
>about that. This is 100% reproducible.
>
>This bug report is probably rather vague, but I probably won't have time
>for anything more detailed or any debugging activities related to this for
>several weeks, and I hope this report is better than nothing.
>
My box also hangs acessing the floppy. Strange thing is that it also
hangs without irqrate-A1. Will send an oops.
--
J.A. Magallon # Let the source be with you...
mailto:jamagallon@able.es
Mandrake Linux release 8.2 (Cooker) for i586
Linux werewolf 2.4.18-rc4-jam1 #1 SMP Sat Feb 23 01:39:06 CET 2002 i686
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
2002-02-23 1:33 ` J.A. Magallon
@ 2002-02-23 8:23 ` Barry K. Nathan
2002-02-23 9:40 ` J.A. Magallon
0 siblings, 1 reply; 11+ messages in thread
From: Barry K. Nathan @ 2002-02-23 8:23 UTC (permalink / raw)
To: J.A. Magallon; +Cc: Barry K. Nathan, Lista Linux-Kernel, rwhron
J.A. Magallon wrote:
> My box also hangs acessing the floppy. Strange thing is that it also
> hangs without irqrate-A1. Will send an oops.
It could be one of the patches that comes before irqrate-A1 in the 00-90
numbering sequence that your patches use; I've definitely reproduced this
without any of the patches numbered higher than the irqrate-A1 patch.(In
my case, if I applied all of those patches except the irqrate one, I
didn't get the freeze. If I applied all the patches up to the irqrate one
after, then I got the freeze.)
-Barry K. Nathan <barryn@pobox.com>
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
2002-02-23 8:23 ` Barry K. Nathan
@ 2002-02-23 9:40 ` J.A. Magallon
2002-02-23 23:39 ` Barry K. Nathan
0 siblings, 1 reply; 11+ messages in thread
From: J.A. Magallon @ 2002-02-23 9:40 UTC (permalink / raw)
To: Barry K. Nathan
Cc: J.A. Magallon, Barry K. Nathan, Lista Linux-Kernel, rwhron
On 20020223 Barry K. Nathan wrote:
>J.A. Magallon wrote:
>> My box also hangs acessing the floppy. Strange thing is that it also
>> hangs without irqrate-A1. Will send an oops.
>
>It could be one of the patches that comes before irqrate-A1 in the 00-90
>numbering sequence that your patches use; I've definitely reproduced this
>without any of the patches numbered higher than the irqrate-A1 patch.(In
>my case, if I applied all of those patches except the irqrate one, I
>didn't get the freeze. If I applied all the patches up to the irqrate one
>after, then I got the freeze.)
>
Looking at the out of sysrq-p:
Pid: 0, comm: swapper
EIP: 0010:[schedule+129/928] CPU: 0 EFLAGS: 00000286 Tainted: P
EIP: 0010:[<c0117651>] CPU: 0 EFLAGS: 00000286 Tainted: P
lf32-i386 -a i386
EAX: 000000ff EBX: c02bac80 ECX: 00000000 EDX: c027c000
ESI: 00000000 EDI: c027c000 EBP: c027dfc0 DS: 0018 ES: 0018
CR0: 8005003b CR2: 4004f6b0 CR3: 1d9fc000 CR4: 000002d0
Call Trace: [rest_init+0/64] [default_idle+0/64] [default_idle+0/64] [rest_init+0/64] [cpu_idle+41/48]
Call Trace: [<c0105000>] [<c0105200>] [<c0105200>] [<c0105000>] [<c0105289>]
Warning (Oops_read): Code line not seen, dumping what data is available
>>EIP; c0117650 <schedule+80/3a0> <=====
Trace; c0105000 <_stext+0/0>
Trace; c0105200 <default_idle+0/40>
Trace; c0105200 <default_idle+0/40>
Trace; c0105000 <_stext+0/0>
Trace; c0105288 <cpu_idle+28/30>
I vote for the sched-O1. The mount never gets rescheduled ????
--
J.A. Magallon # Let the source be with you...
mailto:jamagallon@able.es
Mandrake Linux release 8.2 (Cooker) for i586
Linux werewolf 2.4.18-rc4-jam1 #1 SMP Sat Feb 23 01:39:06 CET 2002 i686
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
2002-02-23 9:40 ` J.A. Magallon
@ 2002-02-23 23:39 ` Barry K. Nathan
2002-02-24 2:24 ` Barry K. Nathan
0 siblings, 1 reply; 11+ messages in thread
From: Barry K. Nathan @ 2002-02-23 23:39 UTC (permalink / raw)
To: J.A. Magallon; +Cc: Barry K. Nathan, J.A. Magallon, Lista Linux-Kernel, rwhron
J.A. Magallon wrote:
> I vote for the sched-O1. The mount never gets rescheduled ????
Just one problem: I can reproduce this without sched-O1.
In fact, all I need is plain 2.4.17 + irqrate-A1. That's it. That's all I
need to get floppy accesses to hang 100% of the time. 2.4.17 without
irqrate does not have this problem.
(I haven't tried getting an oops with sysrq yet.)
-Barry K. Nathan <barryn@pobox.com>
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
2002-02-23 23:39 ` Barry K. Nathan
@ 2002-02-24 2:24 ` Barry K. Nathan
0 siblings, 0 replies; 11+ messages in thread
From: Barry K. Nathan @ 2002-02-24 2:24 UTC (permalink / raw)
To: Barry K. Nathan
Cc: J.A. Magallon, Barry K. Nathan, Lista Linux-Kernel, rwhron
Barry K. Nathan wrote:
> In fact, all I need is plain 2.4.17 + irqrate-A1. That's it. That's all I
> need to get floppy accesses to hang 100% of the time. 2.4.17 without
> irqrate does not have this problem.
For that matter, 2.4.14-pre7 (the earliest that I can apply the irqrate-A1
patch to without rejects) + irqrate-A1 also causes the floppy freeze for
me. By "freeze" I mean my mouse pointer stops moving for me in X and I'm
unable to switch virtual consoles.
-Barry K. Nathan <barryn@pobox.com>
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
@ 2002-02-24 15:00 rwhron
2002-02-24 23:14 ` J.A. Magallon
0 siblings, 1 reply; 11+ messages in thread
From: rwhron @ 2002-02-24 15:00 UTC (permalink / raw)
To: akpm; +Cc: linux-kernel, jamagallon
> > Random Writes
> > Num Avg Maximum Lat% Lat% CPU
> > Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
> > ----------------- --- ------------------------------------------------------
> > 2.4.18-rc2 128 0.67 1.87% 1.334 777.23 0.00000 0.00000 36
> > 2.4.18-rc2-jam1 128 0.80 5.72% 0.190 3.68 0.00000 0.00000 14
> > 2.4.18rc2aa2 128 0.61 1.39% 61.796 72674.58 0.32761 0.32761 44
>
> Holy cow! Are you sure these numbers are right?
Max latency for random writes at 128 threads varies widely. The numbers
above happen to hit on the widest variation.
Random Writes
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
----------------- --- ------------------------------------------------------
2.4.17-rmap12e 128 0.71 5.19% 1.689 981.25 0.00000 0.00000 14
2.4.17-rmap12f 128 0.71 4.97% 0.975 917.19 0.00000 0.00000 14
2.4.18-pre9 128 0.70 1.82% 1.664 785.13 0.00000 0.00000 38
2.4.18-pre9-ac1 128 0.67 1.93% 0.339 398.63 0.00000 0.00000 35
2.4.18-pre9-ac3 128 0.71 5.09% 0.438 951.04 0.00000 0.00000 14
2.4.18-pre9-ac4 128 0.72 5.12% 0.191 11.63 0.00000 0.00000 14
2.4.18-pre9-am1 128 0.76 2.72% 1.527 845.77 0.00000 0.00000 28
2.4.18-pre9-am2 128 0.75 2.83% 1.371 886.19 0.00000 0.00000 27
2.4.18-pre9-mjc2 128 0.71 3.72% 0.220 12.10 0.00000 0.00000 19
2.4.18-rc1 128 0.68 1.81% 1.584 800.21 0.00000 0.00000 37
2.4.18-rc1-slb1 128 0.81 5.77% 0.192 5.25 0.00000 0.00000 14
2.4.18-rc2 128 0.67 1.87% 1.334 777.23 0.00000 0.00000 36
2.4.18-rc2-ac2 128 0.71 5.19% 0.340 555.11 0.00000 0.00000 14
2.4.18-rc2-jam1 128 0.80 5.72% 0.190 3.68 0.00000 0.00000 14
2.4.18-rc4-jam1 128 0.80 5.72% 5.025 6734.19 0.07560 0.00000 14
2.4.18rc2aa2 128 0.61 1.39% 61.796 72674.58 0.32761 0.32761 44
2.5.3-dj5-tio 128 0.75 2.17% 0.209 1.73 0.00000 0.00000 34
2.5.4-dj2 128 0.76 2.23% 0.206 1.83 0.00000 0.00000 34
2.5.4-dj3 128 0.75 2.30% 0.207 1.87 0.00000 0.00000 33
2.5.5 128 0.74 2.09% 0.197 1.97 0.00000 0.00000 35
2.5.5-dj1 128 0.76 2.26% 0.201 1.45 0.00000 0.00000 33
2.5.5-pre1 128 0.75 2.28% 0.197 2.64 0.00000 0.00000 33
> The increased throughput will be thanks to the boosted request
> queue size.
>
> The (greatly) increased CPU load will also be due to browsing the eight-times
> larger request queue. Plus we browse it a bit more than we used to.
>
> The improvement in worst-case latency in both -aa and -jam will
> be due to the FIFO wait for requests.
Thanks for the insight there.
> But improvement by a factor of 20,000 sounds a little excessive :)
> And a maximum latency of three milliseconds would seem to indicate
> that the benchmark is *never* waiting on disk seek, and that
> perhaps the request queue is simply never filling up. But that
> doesn't make sense.
I notice the 2.5.x kernels also have extremely low max latency on this test.
> What does the "latency" actually mean? Is it the time spent
> in the kernel to issue a write(2)?
AFAICT, it's the time between when a request like write(2) is made,
and when it completes. It doesn't appear to be actual time in kernel.
I'd rather hear from a more knowledgeable coder though.
> Something funny is happening, I suspect. Guess I should go
> look at tiobench...
Cool. I admire your work.
Below is a snippet of tiobench on random writes. The rc4-jam1
included the entire patchset, whereas rc2-jam1 had patches with
the first two digits < 20.
Most notable is that jam1 had an unusual low max latency at
16 threads, which didn't appear at 8, 32, 64, and 128.
Random Writes
Num Avg Maximum Lat% Lat% CPU
Kernel Thr Rate (CPU%) Latency Latency >2s >10s Eff
-------------------- ------------------------------------------------------
2.4.18-rc2-jam1 8 0.71 4.71% 0.190 3.53 0.00000 0.00000 15
2.4.18-rc4-jam1 8 0.71 4.61% 0.754 2242.97 0.02500 0.00000 15
2.4.18-rc2-jam1 16 0.72 4.76% 0.192 4.02 0.00000 0.00000 15
2.4.18-rc4-jam1 16 0.72 4.72% 0.196 5.09 0.00000 0.00000 15
2.4.18-rc2-jam1 32 0.77 5.13% 0.192 5.28 0.00000 0.00000 15
2.4.18-rc4-jam1 32 0.77 5.25% 8.268 13672.38 0.15000 0.00000 15
2.4.18-rc2-jam1 64 0.79 5.27% 0.190 3.63 0.00000 0.00000 15
2.4.18-rc4-jam1 64 0.78 5.30% 26.660 27314.64 0.50403 0.02520 15
2.4.18-rc2-jam1 128 0.80 5.72% 0.190 3.68 0.00000 0.00000 14
2.4.18-rc4-jam1 128 0.80 5.72% 5.025 6734.19 0.07560 0.00000 14
--
Randy Hron
^ permalink raw reply [flat|nested] 11+ messages in thread
* Re: [PATCHSET] Linux 2.4.18-rc3-jam1
2002-02-24 15:00 rwhron
@ 2002-02-24 23:14 ` J.A. Magallon
0 siblings, 0 replies; 11+ messages in thread
From: J.A. Magallon @ 2002-02-24 23:14 UTC (permalink / raw)
To: rwhron; +Cc: akpm, linux-kernel
Hi.
First of all, I never expected this tniny patch collection would origin
such "rivers of e-ink"...
Please, have always present that I can have made some mistake in my
offset re-engineering. But the fact that it at least survives the tests
is good.
On 20020224 rwhron@earthlink.net wrote:
>2.4.18-rc2-jam1 128 0.80 5.72% 0.190 3.68 0.00000 0.00000 14
>2.4.18-rc4-jam1 128 0.80 5.72% 5.025 6734.19 0.07560 0.00000 14
This is really strange. I have looked at my patches and are the same. What
changed in mainlaine ??
[...]
>
>Below is a snippet of tiobench on random writes. The rc4-jam1
>included the entire patchset, whereas rc2-jam1 had patches with
>the first two digits < 20.
>
So rc2-jam1 is running without the ide-update (I noticed your system is IDE),
but also without irqrate.
I will reorder the patches so you can apply 0*, 1* and 3* without problems,
then scsi-ide updates. And you can try with/out the ide update.
Do not see what can be related with latency in >=20*, apart from ide and
irqrate...
I do not know if you are already doing this, but I will skip the bproc part
for thins tests. It pollutes system calls with hooks for network, so it can
be hurting in many ways.
I will release a -jam2 with latest vm-27 and reordering the patches.
--
J.A. Magallon # Let the source be with you...
mailto:jamagallon@able.es
Mandrake Linux release 8.2 (Cooker) for i586
Linux werewolf 2.4.18-rc4-jam1 #1 SMP Sat Feb 23 16:25:56 CET 2002 i686
^ permalink raw reply [flat|nested] 11+ messages in thread
end of thread, other threads:[~2002-02-24 23:15 UTC | newest]
Thread overview: 11+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2002-02-22 0:50 [PATCHSET] Linux 2.4.18-rc3-jam1 J.A. Magallon
2002-02-22 7:38 ` Barry K. Nathan
2002-02-23 1:33 ` J.A. Magallon
2002-02-23 8:23 ` Barry K. Nathan
2002-02-23 9:40 ` J.A. Magallon
2002-02-23 23:39 ` Barry K. Nathan
2002-02-24 2:24 ` Barry K. Nathan
-- strict thread matches above, loose matches on Subject: below --
2002-02-22 4:21 rwhron
2002-02-22 5:35 ` Andrew Morton
2002-02-24 15:00 rwhron
2002-02-24 23:14 ` J.A. Magallon
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox