If you track at your chosen rms ref level for each track and you are in 24 bit, no gain adjustments are usually required from that point on.
 
But for whatever reason you may need to add or subtract gain to a track prior to mixing.  (especially and most always if you are mixing a session tracked by someone else. Very few get this correct track rms level thing going on!)  Normalising in that sense just takes the highest peak in the whole file and adds enough gain to the whole track so that peak can reach 0 dB FS.  Which is also not really wanted either.  And it is also a process that factors the peak values in where they are not really important or as important as rms levels.
 
A much better option is measure the rms level of a track and either add or subtract gain to that track  (I usually do that in the editing program)  so that the rms level matches your chosen reference again. This is different to normalisation but a better process for this kind of mix prep.
 
John's approach of starting with nothing and adding stuff in to create a balance is a good one. Even better to that is do this into a stereo VU meter.  The idea is to get the rhythm section pumping to about -3 or -4 dB VU and then by the time you add all the lead elements and other stuff in you have a got a full mix just peaking 0 dB VU.  When all your tracks are at the correct rms level then you can easily attain this situation.  And most often the faders feeding that buss or main buss will all be down a little from unity but almost at a similar setting too.
 
The 0 dB VU ref level remember is down at the digital reference level which may be -14 or -18 or -20 dB FS.
 
That mix will sound huge especially if it has either 14, 18 or 20 dB of transient headroom above the rms level. Turn it up loud in your room and revel in how amazing it sounds.  The rms levels inside your DAW are always related to the SPL level in your room.  When the SPL level in the room is correct you also tend not to push levels up on tracks and buses etc in order to get the volume.  The volume is already there.  In fact when the SPL monitoring level is up then you tend to hold track and buses back a little more and let them breathe etc.
 
We did all this in analog years ago and for me it still applies.  Except then the analog headroom ceiling was a bit like a piece of thin stretched rubber.  Loud things could go up and well over but it was a peak that was absorbed a bit.  In digital our ceiling is more like a thin glass ceiling and we need to make sure whatever we do, still sits a little under it so to speak.  Otherwise we can crash through and the results can be unexpected, usually the signal quality breaks at around 0 dB FS so there is no need to get so close to it really.

I about to setup my sonar system and want to fully understand the meters inside sonar. There will be many pieces of analog gear tied to the system.
 
My converters are designed to take +22DB (in the +4 XLR world) and in theory will relate to 0 FS. They also have very good low level performance unlike the ADATs of old.
 
reading from the loudness alliance and other you tube vids about up and cumming -23 LUFS standard for broadcast. I also have been reading about mixing at K-20. Not sure how they relate yet My question is this.
 
Can I set up the meters in Sonar to work like the old analog meters, were I want things to peak at -3 to +3 DB assuming the equipment has the headroom for +22. And at the same time, be working at say K-20 inside the box? 
 
Let's say there is a way to do that, Now how do I push a plug in compressor and an analog compressor (yes through the converters) the same way, and get all working at the right levels?
 
I have yet to track and mix a single song in cakewalk since I used it on a 386 Dos machine. I have used the software as a host for BFD3, but have only used it for that over the past 5 years. The audio was stemmed and sent to another mixing inside the box I'm more familiar with.
 
Now I want to really dig into sonar, and step one is getting the metering and levels right.
 
If you are wondering some of the equipment specifics: SSL Alpha link AX, MX4 card, Dangerous ST monitor controller, Dorrough 40C meter. Tons of analog preamps, synths, mic's compressors.

If your converters can output a maximum level of +22 dBu then it means at the nominal level eg +4dBu you will need to operate as 0 dB VU as -18 dB FS.  No harm in dropping down to -20 as your reference either.  That just means at -20 the output level leaving your converters will be +2 dBu. Which is still fine.
 
As far as I know Sonar's rms meters are not able to show high up on the scale eg 0 dB VU at the reference level.  The rms meters will always be way down on the scale.  There is also another issue with Sonar rms meters and that is they show 3 dB lower than the industry standard as well.  (they are showing actual rms as being 3 dB down from peak but that is not the agreed industry standard)  The industry standard is that for a continuous tone the peaks of the sine wave should be at the ref level.
 
The best way to overcome the lack of proper VU display in Sonar is to invest in some VU meter plug-ins such as from Klanghelm or PSP.  Both make excellent VU meter plug-ins with a proper VU scale and the ballistics of the meters are close to the real deal too.  They are not expensive either.  You can set the reference level with these plug-ins so if you set the ref to say -20 then the VU will show 0 dB VU at -20 dB FS which is what you are after.
 
http://www.klanghelm.com/VUMT.php
 
http://www.pspaudioware.com/plugins/tools_and_meters/psp_triplemeter/