CS4 jerky AVI playback, ImporterProcessServer?

Question

I usually use v6.5, and AVI playback is fine....Then I did a clean install of CS4, now AVI playback is jerky, choppy, slow...What is up with that?Windows Task Manager tells me Premiere CS4>"ImporterProcess Server" is using over 50% of CPU muscle..is that the problem, whats the workaround?...the PC is plenty fast with enough ram, and has done plenty of successful video editing in many formats(including AVI), in the past
THANKS

Answer 1

Thank you, to both Jim and Harm for a great series of technical posts on this thread.  Really fun to read and learn from!
A study on drive performance I did yesterday on my RAID 0 configuration may add a bit of thought to our discussion.  I evaluated disk access times, and min/max read times.  Plus did reading up about RAID controller cards.
First interesting discovery (at least for me) . . . as your video files get placed deeper on your drive, the access times increase significantly and data rates slow down.  Reports I've read explain this, and it's the simple physics of a platter spinning at a constant RPM.  The read/write head flys over the outer tracks of data much faster than the tracks near the center of the platter.  Sure enough, my read performance graph looks like a dying patient as the test approaches the inner tracks.  Not disastrous, but a good 30% drop in speed.
So first conclusion is that if the drive starts getting close to full, slowdowns can be significant.  A separate plot of access time over position showed that near the end of the drive, access times also increased.  Hard drives start by writing data to the outer edges of the platter, where the head flies fastest and data speeds are at their best.  Data gets placed closer to the middle as you fill up the drive.
Next interesting discovery -- most PC people, like me, use motherboard-based RAID controllers.  Not so good I found out.  Those are all software-based, and use a fair amount of the CPU -- my CPU overhead was around 15%.
With my RAID drive test, I also observed substantial speed "pumping", up and down, which comes from the nature of relying on the motherboard CPU which can't stay focused on just drive access.  My worst-case low points in the pumping got down to about 70 MB/s.
From the reasearch I've done, a PCI-e hardware-based RAID controller will eliminate the pumping and give you around a 20% to 30% speed increase with fairly steady read speeds.  Since my average read rate was about 110 MB/s, I could see that go upwards to 140 MB/s with a hardware-based RAID card, and hold fairly steady performance speeds.  Of course, if video files are fragmented, that'll slow things down somewhat.
I hadn't thought about the effect of multiple tracks of video jumping up the needed bandwith.  A real good point.  But I don't often go beyond 3 tracks of video.  I sometimes do more complex overlays, but then, I will need to render those because my CPU is quickly overtaxed as Jim points out.  And then with rendering, shouldn't the rendered segments be read out as a single video track?  In other words, my workflow would probably rarely involve reading off of more than 3 video tracks, and I'd think Jim's observations are very applicable to me.
So it would seem that RAID 0 gives me quite a bit of performance cushion, even with some of the other system overhead that Harm mentions.
That might change if I went to the professional formats, working with video at 4:2:2, and began moving to less compressed formats.