Probably the widgets you are using are leaking memory. Try removing them and check which of the widget is taking too much memory.
I use Calendar, 2 Clocks, 1 Weather and 2 Webclips and the Dashboard client is taking up 83Mb. Normally I just force quit if I see it is taking too much memory, and start it again if I need to them back.
The obvious thing to try is eject the Time Machine volume (or otherwise make it unavailable - turned off) and reboot the mac. mds will only use RAM when a program is asking it to supply spotlight answers. The files you have are using a large amount of space and being called upon. A clean boot should get your mds with little or no RSS until you start to launch apps or TM starts.
Whether this is normal is harder to tell - even if it's caused by your files, it's likely made worse a direct result of the PPC architecture not having as optimized a time of indexing than the Intel architecture. The mds program likely is coded with parts of the code for PPC and part for Intel as part of the normal performance optimizations Apple or any large corporation will do before releasing a daemon that runs all the time. Furthermore, the frameworks it uses will be native to the processor - so it's not really the same program running when you have a different CPU.
The comments to the question show that you have really down your homework and played around a bit with excluding volumes to see if the large memory is happening in response to many drives being indexed.
Sadly, you don't have any real control over how large the process gets. It simply runs based on the spotlight importers that are present on your system and in response to all the filesystems you expose the spotlight subsystem to.
If the obvious step of isolating Time Machine doesn't reduce the RAM usage, You can also try:
- cleaning out third party spotlight extensions
- add more RAM and live with it
- permanently disable the folders that are causing RAM usage to spike (and forego the usefulness of spotlight)
- temporarily disabling all volumes by putting them on the privacy exclusion list and then slowly adding folders to be indexed and hope to determine which sort of data is causing the worst spike in
- use Shark and their kind (part of Xcode and the CHUD tools) to sample the running mds process and map out the memory allocation as it is happening. This is fairly high on the skill level and won't change your end solution either way - you'll just know why - and maybe still not have a cure.
Also sadly for you - most people are not running 10.5 on PPC chips, so the number of people that can run some tests or have seen this and poked around the issue it to find out what sort of file or importer might be likely to take more memory once mds gets it's hands on the metadata store.
It does sound odd - so you have a very valid point with more than usual RAM usage. But, it's not necessarily something that is wrong - you just might have data that needs that large amount of RAM. Only by taking your data to a similar mac or doing the process of elimination can you know for sure.
Best Answer
Have a look in the Console log. You will likely see that it's a Python-internal issue. Most likely a recursion or some exhaustive memory allocation. Python and other languages use a limited stack/heap to store runtime data (like the call stack). If that is exhausted they simply croak and do not ask the OSX to provide more memory.