bernhold@qtp.ufl.edu (David E. Bernholdt) (03/14/90)
Is there any way in software to determine the number of memory banks on a given machine? Failing that, can anyone point me towards a current table of these values vs. the model, or something similar? I have a rather old table, which gives this info for some of the oldest models, but it doesn't include XMP/EA, YMP, or -2 architectures, and considering the detail, there are probably many more variants which I can't even identify. What other ways could I find this info out for a given site? -- David Bernholdt bernhold@qtp.ufl.edu Quantum Theory Project bernhold@ufpine.bitnet University of Florida Gainesville, FL 32611 904/392 6365
pmk@craycos.com (Peter Klausler) (03/14/90)
parameter (l=11,lp=2048,n=640,nn=n*lp)
integer m (nn), cps (0:l), str
str = 1
do 1 i=0,l
j = irtc ()
do 2 k=1,n
2 m (k*str) = m (k*str) + 1
cps (i) = irtc () - j
1 str = 2*str
print 3, (i, cps (i), i=0,l)
3 format ('2**',i2,': ',i6,:)
end
Run this, examine the output, and find the least power of two with the
worst performance (highest clock count). This should give you a pretty
good idea of the number of banks on the machine. On a multiprocessor this
code will yield different results on each run, but at the stride that
equals the bank count things won't differ by too much.
You can also run the 'target' command (on COS or UNICOS) to see what the
compilers and libraries are supposed to believe about the machine; it's usually
trustworthy.
-Peter Klausler @ Cray Computer Corp. in Colorado Springs, CObernhold@qtp.ufl.edu (David E. Bernholdt) (03/15/90)
In article <949@red.qtp.ufl.edu> I wrote: >Is there any way in software to determine the number of memory banks >on a given machine? Thanks to the several people who replied by mail, along with Peter Klausler, who posted... The target(1) command does what I want -- and provides more than just the number of banks. This command is available in both COS and UNICOS, as Peter pointed out. The manual I have implies that it is not available on a -2, but I don't have access to a unicos -2 to check it out. In UNICOS, the data is obtained from the target(2) system call, which returns the data that target(1) prints out, although the structure isn't documented in the man page. looking at the include file <sys/target.h> shows what the structure is, and so it is possible to easily obtain this info from a running program with a system call. (although the "primary machine type" it gives is a very large integer number, and I don't know how to interpret it -- not that its so important) On COS, I'm sure there is an equivalent system call, but checking the few manuals I have for COS revealed nothing useful. I have no idea what they do in CTSS. Thanks again. -- David Bernholdt bernhold@qtp.ufl.edu Quantum Theory Project bernhold@ufpine.bitnet University of Florida Gainesville, FL 32611 904/392 6365
mjt@duck.ncsc.org (Mike Tighe) (03/15/90)
In article <949@red.qtp.ufl.edu> bernhold@qtp.ufl.edu (David E. Bernholdt) writes: > >Is there any way in software to determine the number of memory banks >on a given machine? Command line: Execute the command called target(1). This will give you the info for your machine. From C program: Make a call to the GETPMC(3U) function. This returns an array of values for your machine. Further details on each method can be found in the appropriate manual page. -- Michael Tighe, mjt@ncsc.org (919)-248-1157
ken@wj.msc.umn.edu (Ken Chin-Purcell) (03/15/90)
In article <950@orange19.qtp.ufl.edu>, bernhold@qtp.ufl.edu (David E. Bernholdt) writes: > The target(1) command does what I want -- and provides more than just > the number of banks. This command is available in both COS and > UNICOS, as Peter pointed out. The manual I have implies that it is > not available on a -2, but I don't have access to a unicos -2 to check > it out. If you mean a CRAY-2, then yes, target(1) does exist. At least it does on our machine (UNICOS 5.0). \/ Ken Chin-Purcell <> also known as ken@msc.umn.edu and (612)626-1340 /\ Minnesota Supercomputer Center, University of Minnesota
lrul00@dixel.Kodak.COM (Richard C. Dempsey) (03/16/90)
In article <950@orange19.qtp.ufl.edu> bernhold@qtp.ufl.edu (David E. Bernholdt) writes: >In article <949@red.qtp.ufl.edu> I wrote: >>Is there any way in software to determine the number of memory banks >>on a given machine? >The target(1) command does what I want -- and provides more than just >the number of banks. This command is available in both COS and >UNICOS, as Peter pointed out. The manual I have implies that it is >not available on a -2, but I don't have access to a unicos -2 to check >it out. On the Cray-2 at NCSA, the target command works: u2 1% target Primary machine type is: CRAY-2 banks = 128 numcpus = 4 ibufsize = 16 memsize = 134217728 memspeed = 35 clocktim = 4100 numclstr = 0 bankbusy = 13 u2 2% I didn't look into the target(2) system call. Presumably it's there if the command is there. Rich Dempsey dempsey@Kodak.COM Computational Science Laboratory (716) 477-3457 Eastman Kodak Company Rochester, NY 14650-2205