| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1184.1 | known problem | TAVENG::BORIS | Boris Gubenko, ISE | Thu Aug 01 1996 03:54 | 100 |
| 1184.2 | the list of the functions | TAVENG::BORIS | Boris Gubenko, ISE | Thu Aug 01 1996 07:39 | 11 |
| 1184.3 | also tzset() and wcsftime() | TAVENG::BORIS | Boris Gubenko, ISE | Thu Aug 01 1996 08:09 | 0 |
| 1184.4 | | CSC32::D_SANFORD | | Thu Aug 01 1996 19:32 | 17 |
| 1184.5 | | TLE::D_SMITH | | Fri Aug 02 1996 07:29 | 5 |
| 1184.6 | s/hopefully/definitely/ | TAVENG::BORIS | Boris Gubenko, ISE | Fri Aug 02 1996 09:34 | 33 |
| 1184.7 | Default to 1 or 2, please? | WIBBIN::NOYCE | Pulling weeds, pickin' stones | Fri Aug 02 1996 10:15 | 4 |
| 1184.8 | I agree | TAVENG::BORIS | Boris Gubenko, ISE | Fri Aug 02 1996 11:28 | 6 |
| 1184.9 | | TLE::D_SMITH | I love Digital, but miss DEC | Fri Aug 02 1996 12:08 | 4 |
| 1184.10 | Ok, I'll change it | TAVENG::BORIS | Boris Gubenko, ISE | Fri Aug 02 1996 14:00 | 0 |
| 1184.11 | Update from customer | CSC32::D_SANFORD | | Wed Sep 11 1996 16:06 | 33 |
| 1184.12 | of course | HNDYMN::MCCARTHY | A Quinn Martin Production | Wed Sep 11 1996 18:52 | 5 |
| 1184.13 | DEFINE DECC$TZ_CACHE_SIZE 0 | TLE::D_SMITH | Let's just say I like Digital alot | Fri Sep 27 1996 08:24 | 17 |
| 1184.14 | Fixed in ovms 7.1? | CSC32::D_BUSTOS | | Wed Jan 08 1997 12:07 | 8 |
| 1184.15 | | TLE::D_SMITH | Duane Smith -- DEC C RTL | Wed Jan 08 1997 13:01 | 5 |
| 1184.16 | localtime() performance OpenVMS vs. Unix w cache enabled | CSC32::D_SANFORD | | Thu Apr 03 1997 16:59 | 30 |
| OpenVMS Alpha V7.1, DEC C V5.5-002
Unix V3.2, DEC C++ V5.4
localtime() is 30 times slower on OpenVMS, with cache enabled (no disk
I/O), then the equivalent function on Unix.
Customer is having a problem with the localtime function taking so long
to run on their OpenVMS V7.1 system that they can't use their software.
They were curious about why they weren't seeing the problem on Unix, so
they ran some comparisons. Theyfound that localtime takes 30 times as
long to run on a VMS machine as on a Unix machine. By comparison,
gmtime takes about the same time on both. Can you please check into
the reason for the difference in localtime and whether it can be made
faster on VMS? We would think they would run about the same on the two
systems.
Here is the data they collected, on Alphastation 255/233 machines.
Each number is the time, in microseconds, for one call to the function
in question. They were collected from runs of 10000 calls each.
localtime gmtime
OpenVMS 583 10
Unix 19.5 9.1
Fortunately, both operating systems and both compilers are Digitals, so
the comparison is, I believe, one of apples and apples.
Regards, Drew Sanford
Customer Support Center
C970227-5970
|
| 1184.17 | Some speculation | QUARRY::neth | Craig Neth | Thu Apr 03 1997 17:26 | 14 |
| localtime() has to do all of the timezone processing. Both VMS and DIGITAL
UNIX use the public domain timezone package, but the DIGITAL UNIX version
has been modified to perform better. (in particular, libc keeps some state
that helps it figure out whether or not it needs to search through all
the tables every time tzset is called) I don't know if those same
modifications have been applied to the VMS version or not.
In the UNIX version, every time localtime is called a getenv("TZ") call
ends up being executed. It could be that this is a faster operation on
UNIX than it is on VMS...
|
| 1184.18 | entered into CRTL internal | TAVENG::BORIS | Boris Gubenko, ISE | Thu Apr 03 1997 21:18 | 10 |
|
We will investigate this. It may be getenv() or another underlying
CRTL function that is called by localtime() and is not called by
gmtime() or, perhaps, VMS missed some performance improvement that
was done in the UNIX version.
Could the customer tell us what time zone was used when running the
comparison?
Boris
|
| 1184.19 | question | CAIRN::HARRIS | Kevin Harris, dtn 381-2039 | Fri Apr 04 1997 14:38 | 5 |
| I studied this once when I was looking at DEC C compile speed for short programs
on VMS. I discovered that it needs to bring in a shareable image to look up
something in the timezone database. I think this is the cause of the long
delay. Is it slow everytime, or just the first time?
-Kevin
|
| 1184.20 | More info from customer | CSC32::V_HAVER | | Mon Apr 28 1997 17:58 | 16 |
| From customer:
...
Yes, it is slow on every call. 10000 calls in a row take 10 times as long as
1000 calls in a row. We were in the PST/PDT time zone. TZ was set to:
"PST8PDT7,M4.1.0,M10.5.0"
When I set the time zone cache to zero (2 is default), localtime became
another order of magnitude or so slower. Setting the time zone cache to 4 or
higher made no improvement over using the default.
...
Thanks,
Vicky (virtual Drew for today)
|