zz:osnews:一些有关内存使用量的摘抄

bbbush · 发表于 2005-5-14 10:46:46

http://www.osnews.com/comment.php?news_id=10555&limit=no

RE: FC-performance
By William (IP: 200.30.100.---) - Posted on 2005-05-11 13:55:23
Hi. I posted a simple performance comparision between FC3 and FC4T2 on d-d-l@gnome.org.
http://mail.gnome.org/archives/d ... 5-May/msg00050.html

The hall of bloat

* From: William Lovaton <walovaton yahoo com mx>

* To: desktop-devel-list gnome org

* Subject: The hall of bloat

* Date: Mon, 09 May 2005 19:39:45 -0500

Hi there guys.  I've been following the Gnome development for a long

time and I am a good advocate of this desktop here in my country

(Colombia).

I write this for two reasons.  First, to congratulate you all for a very

good 2.10 release, it has improvements and refinements all over the

place like every new release.  Second, to give some comparisons results

from Gnome 2.8 and 2.10 about performance and memory consumption using

Fedora Core 3 and Fedora Core 4 Test 2 systems respectively.

The FC4T2 machine (G2.10) is an old Athlon 600MHz computer with 384 MB

of RAM.  The FC3 machine (G2.8) is an Athlon XP 2000MHz computer with

512 MB of RAM. So take the differences into account when reading the

comparison.

First, I configured an standard gnome setup in new accounts, the

standard applets, and a few launchers.  Then I changed the theme for

Clearlooks, the fonts to bitstream and 8 pixels, set a colorful

background, deactivated unneeded services and finally, get rid of the

ugly Redhat Network Applet (rhn) which frees a few megs of RAM.

On both systems the binaries were prelinked to improve start up time and

reduce memory consumption.  Both systems were using the spanish locale.

Here are some results:

FC4T2 FC3

----------- ------------

System boot time 1 min 2 sec 1 min 10 sec

Desktop start up time 18 sec 21 sec

Memory usage after login 62 MB 83 MB

The following chart shows the memory consumption by programs as shown by

gnome-system-monitor.  The first value is the virtual memory size and

the value in parenthesis is the resident size.  Both values are in

megabytes.  The first section of the table will show the processes

started by default when you log into gnome which correspond to the

"Memory usage after login" field in the previous table.

FC4T2 (G2.10) FC3 (G2.8)

------------- ----------

X 66.6 (15.0) 91.1 (18.4)

nautilus 38.2 (13.0) 40.0 (16.0)

g-s-m 32.1 (10.8) 22.9 (10.5)

gnome-panel 31.6 (9.9) 22.0 (10.8)

mixer_applet2 22.3 (9.3) 23.7 (10.2)

wnck-applet 21.1 (9.2) 21.1 (9.3)

gnome-session 20.9 (8.5) 23.4 (9.9)

clock-applet 21.2 (8.0) 21.5 (7.9)

g-s-d 19.8 (7.0) 20.7 (7.3)

gconfd-2 10.2 (6.9) 10.4 (6.5)

metacity 13.3 (6.5) 14.7 (6.6)

g-v-m 19.4 (6.4) 20.1 (6.3)

These were the most relevant processes after login which in theory

represents the 62 MB for G2.10 and 83 MB for G2.8.

The following are the memory consumption for some gnome/gtk+ apps.

FC4T2 (G2.10) FC3 (G2.8)

------------- ----------

evolution 75.0 (18.7) 43.5 (17.2)

firefox 95.3 (23.9) 99.5 (22.9)

epiphany 96.4 (20.4) 99.2 (21.6)

gedit 25.4 (12.2) 24.9 (12.3)

gimp 33.4 (26.3) 31.5 (23.2)

evince/gpdf 23.4 (9.7) 22.3 (9.2) [1]

gconf-editor 20.8 (9.3) 20.8 (8.8)

dia 27.7 (14.9) 27.6 (14.8)

planner 25.0 (11.9) 34.5 (14.8)

totem 35.4 (13.5) 121.0 (30.6) [2]

glade-2 23.3 (12.1) 23.7 (11.9)

gnome-terminal 35.7 (11.7) 37.0 (12.0)

gcalctool 21.0 (9.3) 21.1 (9.1)

------------- ----------

[subtotal] 151.3 MB 181 MB

oowriter 152.9 (51.3) 146 (52.8)

------------- ----------

[total] 167.8 MB 199 MB

[after closing

all apps] 81.1 MB 99 MB

[1] evince for FC4 and gpdf for FC3

[2] FC4 uses the GStreamer backend, FC3 uses the totem backend.

These values were obtained just opening the apps without loading any

file or document, eg. firefox and epiphany were configured to load a

blank page by default.

During constant use it was a lot easier to raise memory consumption on

the G2.8 system.  G2.10 was a lot more consistent and it was very

difficult to make the system swap.  For this to happen I had to use

several apps and load many documents (word, pdfs, images, etc.).

You can see that nautilus have improved over the previous version, the

same goes for X and gnome-session.  gnome-panel got a reduction in the

resident size bot got bigger in the virtual memory size.  What does that

mean? What would cause this change in behavior?

I know that the new version of evolution have got some optimizations

that would reduce its memory usage over the previous version when

working with a huge mail box.

The other doubt I have is the difference of memory consumption when I

started the desktop (62 MB), started many apps, and closed all the apps

(81.1).  The difference is 19 MB for G2.10 and 16 MB for G2.8.  What

would that mean? any leakage problem?

The other problems I see are the applets, the clock for example, gets

bigger after using the evo integration going from 21.2 (8.0) to 64.0

(10.4) of virtual and resident size respectively, even evince is

smaller.  The reason for this is that the clock is not simple any more

because it got new features and it integrates very nicely with

evolution.  Any idea about this?  I know Ben Maurer have been suggesting

to change some applets in-process (into gnome-panel) to reduce the

overhead but this is not a very good idea stability wise.

A suggestion of mine (I'm not a C expert) would be to move some applets

(or group them) into some programs, for example, the trashapplet uses 9

MB of resident size which is a lot for such a simple functionality.

Maybe we can save a lot moving the trashapplet into wnck-applet or

creating a new program that would group some applets which have common

functionality.  That way we can get the best of both worlds, a very

stable panel and less independent processes running on the system for

every single applet.

To summarize, we can see a lot of improvements in this new Gnome version

and I am looking forward to the optimizations that will be in G2.12.

I'd really like to see the memory consumption of the standard desktop

reduced from 62 to less than 50 MB.

The other thing I'd really like to see optimized is the desktop startup

speed, like Nat Friedman said in an earlier thread, it takes a very long

time.

Cheers,

-William
复制代码

In short, FC4 is a lot better in memory usage. More that 20 MB less that FC3 with an equal desktop setup. I turned off unneeded services with out being paranoic (I left sendmail turned on I swear :-)).

The boot up process is faster too (I guess I wasn't using early login) and the Gnome login speed seems better. All of this with a not so fast computer, FC4 was on an Athlon 600 while FC3 was on an Athlon 2000.

About xfs, I just checked out top in my FC3 system and it says 3.5MB of VM size and 1.7MB of residen size. That is not a big deal. If you want to save memory you should kill the RHN Applet for good... lots of megs in there.

About SELinux, I just noticed that I consumend a bit more of RAM but everything was fast. During the comparision I posted I turned off SELinux but that's something that everyone should have enabled all the time.

Finally, remember that the most important thing is to use prelink after installation. It can take about 1H 30Min depending on the number of packages installed. It really makes a difference in start up speed and memory consumption.

Cheers,

-William

2.

RAM Usage
By Father Baker (IP: ---.lcwireless.net) - Posted on 2005-05-11 13:19:11
Why do people think that filling RAM is a bad thing? As long as you're not heavily swapping, you *should* be using your RAM. You paid for it, remember? Having 512 MB of it sitting idle while programs starved themselves would not make your system "optimized."

RE: RAM Usage
By William (IP: 200.30.100.---) - Posted on 2005-05-11 14:08:43
Father Baker, It is _always_ good that a developer try to reduce memory usage of a program (without hacks) because of the reason you just said.

The RAM shuold always be full, like you said, but with the OS buffer and cache data, _NOT_ with user space software bloat. That's the big difference. When you read files or do other operations the OS will cache as many data as it can as long as it has spare RAM. If you start many programs, or use a lot of bloatware, the OS will have to reduce the cache size to give memory back to user space programs.

Another advantege is that smaller programs improve cache behavior on the processor and reduce cache misses making the system more responsive.

Cheers,

-William

3.

RE: RAM USAGE
By JeffS (IP: ---.ded.pacbell.net) - Posted on 2005-05-11 17:12:12
"Why do people think that filling RAM is a bad thing? As long as you're not heavily swapping, you *should* be using your RAM. You paid for it, remember? Having 512 MB of it sitting idle while programs starved themselves would not make your system "optimized.""

Bingo. I just learned this myself recently. I was always getting anal about Linux RAM usage. If I had it running on a machine with 128 megs, it would always be using around 124 megs. If it were on a machine that had 256, it would be using around 248. On a machine with 512, 490, and so on. So this led me to believe that Linux was being inefficient with RAM. I was completely wrong.

This is because the Linux kernel uses RAM intelligently. I caches data files that programs use into memory, since reading from memory is a zillion times faster than reading from hard drive. This speeds up performance of the applications.

Then, when another program, say a big one like Mozilla/Firefox, is launched, instead of going into swap (as one would expect since it appears that it's at near capacity already) the kernel first goes to the unallocated space, then deallocates the cached data files and reallocates those to the new program. And if the data cache file has been changed, the kernel will "flush" it to the hard drive (saving changes, of course). If the data file has not been changed, the kernel will simply deallocate it.

This RAM data file caching gives the best of all worlds. It speeds up the performance of apps, and efficiently allocates RAM for all apps.

Just type "free" at the command line. It will report total capacity, total RAM usage, buffers, cache, +/- buffers/cache, swap capacity, and swap usage. The total RAM usage does not reflect what the kernel, the services, and the apps are actually using themselves, but what they are using plus the buffers and cache. The actual usage is the +/- buffers/cache figure.

Windows, by contrast, doesn't do this caching technique (AFAIK), and actually uses lots of memory for the apps and itself (that's not buffers/cache). Someone who really knows internals of Windows please feel free to correct me if I'm wrong.

So with this knowledge, if the "total" RAM usage is not near capacity when you type "free", then you should actually be concerned because that would mean the kernel is not using memory in the most efficient manner possible. But I've never encountered this myself.

In short, Fedora, as well as most other distros, is actually quite efficient with RAM usage. Making it faster or use less memory is a matter of turning off unneeded services and/or depricating features.

4.

yeah, intelligent... right...
By Steve (IP: 12.178.38.---) - Posted on 2005-05-12 10:09:55
When I said Fedora's performance was poor in comparison to other things I didn't mean noticable surface performance, I meant underlying i/o performance.. base install it's just slower at many "common" user activities, many things by only a few seconds, a frighteningly sizable sum by several minutes (I'm running pretty small tests first, on a scale of something like full system sources the differences could be a lot more, up to 30-90 minutes, who knows, I haven't tested anything that large yet due to current time constraints, but I'll get there).

The one big difference I've noticed is bzip2 though. For whatever reason bzip2 in Fedora was seriously about 15-25% faster than in yoper, slackware, arch, etc. Not sure if it's just some compile option redhat used or what, but aside from that one outstanding case it was pretty sluggish on everything else... after I get a base install data set on everything I'll try recompiling it myself from clean sources to see if that is in fact what the difference is.

Anyway, what I was saying about memory usage is not that "low memory usage means a system is optimized", only that low memory usage comes from cutting the frills out. You don't need 15 extra plugins compiled in/loading with Gaim, for instance. At least I don't. ispell is about the only thing with it I use, anything else only wastes RAM and increases the program load time. There seems to be a lot of that, though I don't have the charts handy just now to say exactly what the differences are.

Any attempt at a "One size fits all" approach ends up with bloat, some in HDD space, some in memory, in the case of Fedora it just seemed to be a bit of a mix of each.

Normal users probably won't really see a difference unless they start going crazy with leaving things open like I do, but I'm not really even concerned with that.

I've seen too many "Well that was pretty boring, why doesn't anyone ever go into the internals" comments on here, so that's exactly what I'm doing... examining I/O performance, standards compatibility, available software, ease of porting simple applications, a 30000 foot view of the internal design ideas, available performance tweaks, time install/setup takes to get to a workable system, general usablitily, etc... which is taking forever, as I'm not even done getting the ISOs yet for everything.

Any suggestions on tests anyone would like are welcome at this point also, as this is going to be my summer project, taking about 3-4 days with each system to give a full summary of strengths and weaknesses.

5.

exactly what I said...
By Steve (IP: 12.178.38.---) - Posted on 2005-05-12 10:39:04
"In short, Fedora, as well as most other distros, is actually quite efficient with RAM usage. Making it faster or use less memory is a matter of turning off unneeded services and/or depricating features."

Which is why I said "It's tweakable if you want to spend the time."

I mentioned "base install" specifically...

But given that it's install takes about 5x as long as a lot of other things, plus having to tweak it to get it to where I want, it's not for me...

Also, 300MB of memory usage at fresh boot is not "Intelligent use of memory" no matter what. Not when the same kernel and userland systems respond faster on another system with 1/2 the memory usage. In this case it was mere bloat (read: mostly was not cached memory). It was also a test run with FC4RC3, or whatever it was (it's late, I don't feel like looking), they may have cleaned it up a bit.

But the point I was trying to make still remains. If I boot a system and it is using the bare minimum of what I need, instead of trying to make sure it can handle every weird thing I might choose to do, then there is more memory available by default. That's more memory for a system cache, more memory for paging crap it thinks I'm going to use but won't, more memory to keep me from waiting while it decides to swap it's cache at the same time it tries to load firefox for me.

And that argument goes both ways... "The main difference between FreeBSD and Linux in this regard is that FreeBSD will proactively move entirely idle, unused pages of main memory into swap in order to make more main memory available for active use. Linux tends to only move pages to swap as a last resort. The perceived heavier use of swap is balanced by the more efficient use of main memory."

Yeah, that means in your "high cache" machine that when you load up firefox you have to wait for two things. First you have to wait for a disk swap to free up some of your "quick access cache", and second you have to wait for firefox to actually load into memory. That's twice as much disk activity that you have to wait for at that exact time...

Which makes the cache a good idea only about 50% of the time, which means more "live" memory usage (non-cache) increases the chances of causing the "cache-and-load" sequence to occur... and Fedora, "base install" has a much higher "live" memory usage than other things I've tried.

The "intelligent" memory usage of any system is only a best case idea.

I'm actually going to be doing a bit on Windows 2003 just for comparison sake... in response to your "Windows doesn't do this caching thing (AFAIK)" comment... on this current W2K3 install I have exactly 237MB taken up by the system cache... yes that's right, the cache. Every developed OS has some form of the caching scheme, though the ideas behind them vary a bit.

Windows does like paging a lot more than other things though... or so it seems from the surface. I have to wait on firefox to reload from the page file even though I have 200MB of completely free memory at the moment (minimizing tends to == swapping it seems, I'll look into that more though).

		自动登录	找回密码
密码			注册