| T.R | Title | User | Personal
Name | Date | Lines |
|---|
| 1851.1 | more informatin | SEDSWS::SIAREY | | Tue Feb 11 1997 10:16 | 8 |
| Customer has just called me again to give some more information on this,
it is actually the real time clock drift that the customer is interested
in not the system clock.
Also do we know the name/manufacturer of the clock chip.
Manh thanks,
Colin S.
|
| 1851.2 | 1 minute per month | WRKSYS::SCHUMANN | | Tue Feb 11 1997 14:24 | 7 |
| I believe the AS600 uses a Benchmarq BQ3287 or similar part. The accuracy spec
for that part is "+/- 1 minute per month at 25 degree C." Other similar parts
are likely to have similar specs. This is the clock that Unix uses to do
1ms interrupts. The CPU clock (i.e. RPCC) uses a separate time base that is
less accurate.
--RS
|
| 1851.3 | 1 minute per month | WRKSYS::MURRAY | | Tue Feb 11 1997 17:25 | 1 |
| The AS600 uses a Dallas 1287 toy clock. It is accurate to +/- 1 minute per month
|
| 1851.4 | See AARM For Specified Drift, OS Can Drift Further | XDELTA::HOFFMAN | Steve, OpenVMS Engineering | Wed Feb 12 1997 09:58 | 5 |
|
The worst-case drift is quoted in the Alpha architecture specification,
and in the Alpha Architecture Reference Manual. The running clock can
skew further depending on memory errors or other local high-IPL activity.
|
| 1851.5 | similar question | SEDTMA::SIAREY | | Wed Feb 12 1997 12:42 | 18 |
| Many thanks, for the answers so far, I have given these to my customer
and he is uncertain if this is what he needs so he has defined his
needs further.
If he uses a POSIX4 "clock_gettime (CLOCK_REALTIME,xxx)" and then 10 secs
later issues the same system call, what is the worstcase drift over
this period.
They know that when using this function call they see a resolution of
976 micro seconds, but need to know the worst case drift.
Please excuse ignorance of the question, this is as the customer has
asked me, if this is not the correct conference please could someone
point me in the right direction.
Many thanks for your help,
Colin S.
|
| 1851.6 | 500 microseconds | WIBBIN::NOYCE | Pulling weeds, pickin' stones | Wed Feb 12 1997 14:36 | 8 |
| From the Alpha Architecture Reference Manual, section II-A, page 6-22:
The accuracy of the interval clock must be at least [sic]
50 parts per million (ppm).
So in 10 seconds, the clock can drift no more than 500 microseconds.
You need to add this to the quantization error caused by the ~1msec
resolution.
|
| 1851.7 | It's a computer not a chronometer | BBPBV1::WALLACE | john wallace @ bbp. +44 860 675093 | Thu Feb 13 1997 04:01 | 7 |
| All of this is irrelevant if the customer's system has a clock fault
(it could happen). For known good time, use a known good external time
source - e.g. a GPS receiver with a serial interface. Other approaches
are possible.
regards
john
|
| 1851.8 | Question answered - thanks | SEDSWS::SIAREY | | Fri Feb 14 1997 04:31 | 8 |
| Many thanks to you all for the replies to my question.
I have given the customer all the information and he is happy.
Another success for the notesfiles.
Best regards,
Colin S.
|