Re: linux-image for i586
Arnt Karlsen <arnt@c2i.net> writes:
> On Fri, 12 Nov 2010 16:57:09 +0700, Sthu wrote in message
> <[🔎] 4cdd0f80.ce7c0e0a.2046.ffffa1a6@mx.google.com>:
>
>> Thank You for Your time and answer, Arnt:
>>
>> > > What are the hardware specs of the machine you have? Exactly
>> > > which CPU, how much RAM, what hard disk (size and rpm), etc.
>> > > Please provide the following output:
>> > >
>> > > cat /proc/cpuinfo
>> > > lspci
>> > > free -m
>> > > dmidecode
>> > >
>> > > (If your machine is very old, dmidecode may not work).
>> >
>> > ..'top|head -n 20 ' to see what's eating your iron.
>>
>> Could You please tell me what is for? What do we want to conclude from
>> the given info - that will be a lot - speaking about the last utility?
>
> On Wed, 10 Nov 2010 18:41:26 -0500, Johan wrote in message
> <[🔎] 87aalgiw61.fsf@emmy.axel.nom>:
>
>> The i486 kernel will run on an i586 box. Even if you compile the
>> kernel for i586 how much speed are you really going to get? A few
>> percent? And how much time do you spend in kernel as opposed to
>> userland? Maybe 3-20 percent depending on what you are doing right?
>> So, in rough numbers, that's a 10% off on 10% of what you do for a
>> net of roughly 1%.
>
>
> ..it depends, 'top |head ' tells us another wee story: ;o)
Correct me if I am mistaken.
Looking at your numbers, it seems that most of the time is spent in
userland and not the kernel. I take it that the "us" field represents
time spent in userland and "sy" is kernel, but you probably can also add
in "wa" wait and "hi", "si" to handle interrupts. "st" is, according to
man top, stolen by other like hypervisor.
In the first case, you are 96% in user and only 4% in the rest. In the
second, top reports 60.7% user and 38.6% idle, and the rest sums to
under 1%. So, making the kernel faster would really only help you by
upto 4%.
Ideally, the kernel should be zero (although that is clearly
unrealistic). Some tasks like running a serial port or a heavily used
network filter can cause more kernel activity. Usually, I find that the
kernel is mostly out of the way.
You could gain significantly by speeding up your useland process as that
is where, quite rightly, most of you CPU is going.
> top - 21:36:01 up 11 days, 20:14, 1 user, load average: 1.08, 1.02,
> 1.01
> Tasks: 57 total, 1 running, 56 sleeping, 0 stopped, 0 zombie
> Cpu(s): 95.9%us, 0.8%sy, 0.0%ni, 1.5%id, 0.2%wa, 1.3%hi, 0.3%si, 0.0%st
> Mem: 155764k total, 149516k used, 6248k free, 8400k buffers
> Swap: 0k total, 0k used, 0k free, 111296k cached
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 12009 root 20 0 51868 15m 1612 S 89.1 10.0 952:08.72 motion
> 12691 arnt 20 0 2320 1004 780 R 5.0 0.6 0:00.08 top
> 1 root 20 0 2024 488 396 S 0.0 0.3 1:09.26 init
> arnt@bacdor:~$
> ..the cool breezy competition:
> top - 21:49:26 up 11 days, 20:13, 1 user, load average: 0.36, 0.38,
> 0.36
> Tasks: 60 total, 1 running, 59 sleeping, 0 stopped, 0 zombie
> Cpu(s): 60.7%us, 0.2%sy, 0.0%ni, 38.6%id, 0.1%wa, 0.4%hi, 0.1%si,
> 0.0%st
> Mem: 92336k total, 76504k used, 15832k free, 4772k
> buffers
> Swap: 265032k total, 2696k used, 262336k free, 47080k cached
>
> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
> 22145 root 20 0 49820 13m 1692 S 42.1 14.7 697:25.47 motion
> 22966 root 20 0 2324 1036 808 R 1.9 1.1 0:00.04 top
> 1 root 20 0 2028 236 208 S 0.0 0.3 0:35.17 init
> cm46:~#
--
Johan KULLSTAM
Reply to: