> Are there any performance counters such as other processors > have (cache > hit/miss, tlb hit/miss, branch predict misses, insns retired etc.) > available from Linux?
Hmm. If gcc generates the code to collect them (or gprof), then they should be available in the same ways as on the other platforms. gprof support isn't all that mature for the 390, so it may not work correctly yet. I haven't tried this in a while, so don't have a good answer for you now, but there's the weekend...8-) > > >, machine = 9672" > > > > Model of the box. > > OK, so *any* way to tell which type of CPU type (G3, G4, G5...) I'm > running on? (the nice people letting me use this guest are on > vacation, > and I hate waiting for them to get back :-)) With just the CPUID, I don't think so. The CPUID doesn't contain that information -- you might be able to guess from the serial number, but I wouldn't depend on the answer very much. I also probably generalized a bit too far; it's really more of a processor architecture family than a model number -- up until the IFLs and IEEE FP support showed up, it really didn't matter much, and the processor family type doesn't change with model numbers (ie, it's always a 9672-mumble, and the mumble tells you how many processors of what type are installed, eg a 9672-RA5 and a 9672-RB5 are G4s, both are 9672 family architecture, and the only difference is an RB5 is 2 engines and a RA5 is one engine). The Z boxes are a different architecture type (2064), but that probably doesn't help you much. > > > -- Cache sizes and speeds (bw and lat) for the different levels of > > > cache? (I could use lmbench, but I only have access to one CPU) > > IBM Journal of R&D occasionally publishes some interesting > articles in this > > vein. > Thanks, that paid off immediately, if someone wonders the URL is: > http://www.research.ibm.com/journal/rd > And they have many nice articles online. IBM generally doesn't publish stuff like this, but the R&D journal is usually one of the few places such data gets published if it ever does. It's also a really nice window into the thousands of areas in which IBM does basic research. Not easy reading (the math can get pretty hairy), but worthwhile. -- db
