Running these tests gives some information, but it is probably a little hard to interpret. On a fresh boot, sage will take roughly 18 seconds to start up on my machine. Subsequent runs, however, take roughly 1.8 seconds, typically.
This is all dependent on many things that the operating system does to cache data read from disk. If you have plenty of ram, and it isn't being used for anything else, then the operating system won't actually have to read anything from disk the second or third time you start sage, which explains why there are wildly varying startup times. (Maybe everyone is aware of this already.) So, it would be nice for both numbers (18 seconds and 1.8 seconds) to be smaller, but they can't really be compared. The first depends mostly on hard drive speed while the second is bound by the speed of CPU and by memory bandwidth/latency, and perhaps by CPU cache size/speed as well. On Wed, 2010-03-03 at 17:16 +0000, Dr. David Kirkby wrote: > Ticket http://trac.sagemath.org/sage_trac/ticket/8254 > > "sage takes way too long to startup" > > seems to irritate a lot of people. It does not me too much, but I feel one > way > to at least start to tackle this probably is to get some quantifiable data > and > see where the time is being spent. My hunch is that this is due to disk I/O. > > If we all append our results running > > time echo "2+2;" | /absolute/path/to/sage > > where /absolute/path/to/sage should be a *local* file system. If it is shared > via NFS, then it just adds another unknown. > > Then we might get somewhere and see if there is a common problem, though I > think > 'dtrace' on Solaris or OS X is likely to be the most useful tool for > debugging > this. > > Considering the age of my SPARC (it is circa 2000), I'm getting a startup > time > less than some are reporting with what I suspect is much more modern > hardware. > One thing my SPARC does however have is disks and a disk interface which is > probably superior to 90% of modern PCs. > > I'd suggest running it 5x, reporting the median value for "median" values of > "real" time, and as well as maximum and minimum. > > So here is one set of data/ > > 1) Sun Blade 2000, circa 2000 > 2 x 900 MHz UltraSPARC III+ CPUs > Load average 1 (Sorry, I'm doing something and can't stop that) > 1x 147 GB Seagate SEAGATE-ST3146807FC. 15,000 rpm SCSI with a 2 Gbit/s fibre > channel interface. > Sage version 4.3.3 with patches for Solaris as documented at > http://trac.sagemath.org/sage_trac/ticket/8409 > Median of the 5 runs is: > sage: sage: > > Exiting SAGE (CPU time 0m0.17s, Wall time 0m0.19s). > > real 0m8.263s > user 0m6.502s > sys 0m1.392s > > Maximum time of the 5 runs is > real 0m8.356s > > Minimum tie of the 5 runs is > real 0m8.066s > > I can't be bothered to determine whether that is statistically significant or > not, but it seems to me that data is pretty reproducible. > > What is clear here is that this seems to be CPU bound on my old machine, > which > is hardly surprising given the old CPUs, but 15,000 rpm disks. That might > give > us a clue. > > Anyone else care to put their data here too? > > Dave > -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
