On Sun, Mar 10, 2013 at 01:18:05PM +0100, Seth wrote:
> I like the fact that cache/buffer is normally shown, as (a) it reminds 
> people how block, dentry and inode cache work (b) it shows me whether my 
> system's memory is being utilized.
> 
> Remember we're not all sys-adminning virtual server hosts.

Well, you don't have to be a sysadmin to find it useful. As you put it,
they generally use other, more detailed tools with reports over time.
What I really find useful about the patch, as a user (ok, a power-user,
but people caring about either buffers/cache or dirty/clean all are),
is that I can test a use case and see a live, global state of the system
in a small corner of the window at the same time. It can help, e.g.
noticing some linking operations take a lot of memory in a build
process, or tune a virtual machine by giving it less memory and making
it rely on the host cache instead. When my user interface suddenly
starts to freeze, I can readily know if it comes from writebacks or
something else.

Now, I agree a figure in the 10M is really low. And that's a concern.
I think it comes from the strong limit put on the dirty memory limits,
but it may also be caused by internal LVM operations, I don't know.
What I know is that, with the tuning I did on my machines, It's not
rare to have several hundreds MB (even thousands on rare occasions) of
dirty memory (e.g. when creating file system images for some embedded
devices. While some people might argue that it's insane to have such
amounts of dirty memory, if either the data isn't that significant
(i.e. you can recover it easily or it's simply volatile), or if you
rely on your hardware to be fault tolerant (no unnotified power cut),
and your working set fits in the page cache, it just boosts performance
a lot.

About the fact that it's volatile, again, I have to disagree. It's just
as volatile as cpu usage, number of tasks or load average, which are
what top/htop already show. And the default lifetime of 30 seconds
means that one of the main use cases (machine gets slow, let's start
htop and see what's happening) can perfectly report the values in a
useful way.

Also, showing buffers/cache doesn't say anything about dentry and inode
caches... The patch merely merges the buffers and file cache into a
dirty/clean unified page cache. The sum of buffers+cache should be the
same as dirty+cache, so it continues to show whether memory is being
used.

What do others think about it ?

-- 
Richard Braun

------------------------------------------------------------------------------
Symantec Endpoint Protection 12 positioned as A LEADER in The Forrester  
Wave(TM): Endpoint Security, Q1 2013 and "remains a good choice" in the  
endpoint security space. For insight on selecting the right partner to 
tackle endpoint security challenges, access the full report. 
http://p.sf.net/sfu/symantec-dev2dev
_______________________________________________
htop-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/htop-general

Reply via email to