Re: Solaris 2.6 high pages scans

From: Darren Dunham (ddunham_at_redwood.taos.com)
Date: 05/17/04 

Date: Mon, 17 May 2004 15:32:17 GMT

Mothra <mothra@mothra.com> wrote:
> I have a Solaris 2.6 server (E6500) with 11GB main memory and 11GB swap
> and have recently noticed (using sar -r) that the page scanning daemon
> is consistently running at around 3,200 scans per second. I can't
> account for this (unless I'm missing something) because if I run sar -g,
> I can see the the amount of free memory rarely dips below 230MB and the
> lotsfree param is 180MB.

Do you have priority_paging enabled? If so, it would kick in at
CACHEFREE, which should be LOTSFREEE * 2. It could also be that the
scanner is predicting the amount of memory that will be left in the
future. Is the 'de' field non-zero?

> The only thing I can think of is that sar -g (like vmstat) is vastly
> overestimating the amount of free memory left - I seem to remember that
> before Solaris 8, it does not take into account the size of the segmap
> cache when making this calculation.

> Does this mean that the high page scanning rate does not necessarily
> mean I'm short of memory?

In Solaris 7 and earlier, the disk cache is accumulated from the free
list and is seen by the page scanner as in use. This causes memory
pressure. In other words, all disk I/O in 7 and earlier must be sent
into the free list, which must be enlarged by the page scanner. Because
of this, you cannot (easily) separate the effects of high I/O (normally
good) with insufficient RAM (normally bad). Both cause high scan rates
(which affect performance).

Because of the way the scanner works, the more RAM available, the more
work that it has to do to actually free memory.

Not until 8 is the file cache redone so that it does not induce memory
pressure. On a stable system, I/O causes no page scanning at all. This
can be a huge performance win at high I/O rates. It also means that
page scanner operations are a more decisive indication of insufficient
RAM. 

-- 
Darren Dunham                                           ddunham@taos.com
Senior Technical Consultant         TAOS            http://www.taos.com/
Got some Dr Pepper?                           San Francisco, CA bay area
         < This line left intentionally blank to confuse you. >

Relevant Pages

  • Re: RFT: updatedb "morning after" problem [was: Re: -mm merge plans for 2.6.23]
    ... memory freed it so there is free memory that could be filled with data ... in some other situations swap prefetch cannot help becouse useless data is getting cached at the expense of useful data. ... the design is to use CPU and I/O capacity that's otherwise idle to fill free memory with data from swap. ... Upon completion, it actually frees enough memory that swap-prefetch _could_ help on some boxes, while the real issue is that they should first and foremost dump GNU locate. ...
    (Linux-Kernel)
  • Re: E6500 low on memory ?
    ... > I have a Sun E6500 that I think needs more memory and while I have ... > file cache utilising any free memory. ... The rest can be used by filesystem ... cache etc and it's freed when necessary by a scanner. ...
    (comp.unix.solaris)
  • Re: [patch 02/11 -mmotm] lowmemorykiller: Dont count free space unless it meets the specified li
    ... I'm not seeing what this patch has to do with contiguous free memory. ... kernel freed cache memory (since the lowmemorykiller looked at the sum ...
    (Linux-Kernel)
  • Re: ARM assembly
    ... on conservative mark/sweep. ... the VM itself tries to free memory ... memory when he knows that it is no longer needed. ... where there is capture but the variables are marked final (and captured ...
    (alt.lang.asm)
  • Re: troubleshooting slow disk writes on squeeze.
    ... consuming memory. ... Blue and Yellow show memory used in buffer cache. ... very little memory used and a lot of blank space showing free memory ...
    (Debian-User)