Robert Wolfe (rwolfe++at++dciem.dnd.ca)
Mon, 03 May 1999 21:30:02 -0400
I am having some difficulty displaying let alone manipulating
a very large NFS-mounted (over Fast Ethernet) openflight terrain
database (~1G over 273 files on disk with a total of 1.7M triangles)
with perfly, first on a single processor IRIX 6.2 Onyx IR with
128M RAM and now on a dual processor IRIX 6.4 Onyx2IR2 with 256M RAM.
The first error I experienced during loading on the Onyx
was the following supposed X error which seemed to disappear
when I raised system parameters rlimit_rss_max and rlimit_rss_cur
and added swap resource.
Question 1: Out of curiosity, any idea
what this error is about ?
3151 PF Notice(2): level FP Debug (6) set by environment
variable PFNFYLEVEL
3151 PF Info: PFSHAREDSIZE overriding size to
1243887.70 KBytes 3151 PF Debug/Resource: pfInitArenas() Allocated
1243887KB: 0x10040000 - 0x5befbec8
3151 PF Debug/Resource: pfDataPool::create() - pid 3151 - 16384
bytes at 0x5befc000 succeeded for /usr/tmp/pfClock.3151.pfdpool
3151 PF Debug: Using CYCLECNTR clock (32 bits at
0xb8020000)
3152 PF Debug: Clock is 32 bits with 21.00ns resolution.
3152 PF Debug: Clock wrap handler runs every 1.00000
seconds.
3151 PF Debug: pfOpenWSConnection - opening shared
display :0.0=0x10035320 for pid 3151.
X Error: BadMatch
Request Major code 142 (GLX)
Request Minor code 5 ()
Error Serial #18
Current Serial #18
Perfly did finally display the terrain on the Onyx after 3 hours of
loading,
but promptly crashed demanding more memory/swap space when I tried to
manipulate
the viewpoint. Not too surprising...
I moved on to the Onyx2 hoping to have enough resources there. With
twice the
RAM and more local swap space (over 1G, /dev/zero = 512M), I still
cannot manage
to increase the maximum shared memory arena size to beyond about 512M
using
environmental variable PFSHAREDSIZE even though the theoretical limit is
1.7G,
and despite the fact that I was able to raise it to 1.2M on the Onyx
(see above)
which has lower values in every resource category.
rlimit_rss_max and rlimit_rss_cur are both over 1.5G. During loading,
perfly
crashes again demanding a larger shared arena.
Question 2: What is limiting PFSHAREDSIZE and what are the implications
for
a large dataset?
Question 3: Assuming that I succeed in loading the database on the
Onyx2, what
optimizations will bring me significantly closer real-time performance ?
Perfly is already gobbling up over 60% CPU cycles at peak working times.
It seems that one such optimization for such a large area database would
be
database paging.
Question 4: From this openflight database, is there a way of generating
a database format with built-in paging that is compatible with Performer
?
Would using pfconv to produce a .pfb format work?
Is there a similar benefit when the performer Openflight loader converts
CAT nodes to pfASD ?
Thanks for any assistance,
Robert
-- Robert Wolfe, consultant Simulation and Training Defence and Civil Institute of Environmental Medicine 1133 Sheppard Ave., W., Toronto, Ontario CANADA M3M 3B9 (416) 635-2065 robert.wolfe++at++dciem.dnd.ca
This archive was generated by hypermail 2.0b2 on Mon May 03 1999 - 18:30:10 PDT