---------------------------------------------------------------------------
[Overview]
Our thanks got out to all the people who responded. Please take note
of who contributed in the [Thank You] section. In my original query
we were wondering what effects synchronizing time would have on a
living database. We wanted to know specifically if there were any
gotchas we should worry about. In the [Summary] section you will find
a selection of answers to that question. My original [Q] query
section is also attached to the bottom of this document.
Ohhhh, and just skip over the [History] section. I have reverse
writer's block right now, but you don't have to burden yourself with
it.
----------------------------------------------------------------------
[Thank You]
I received responses from the following individuals. A million thanks
go out to those techno-warriors who help me, and may they keep helping
too because I don't forget to acknowledge them, but may a pox fall on
those individuals who do! :)
roy_at_endeavor.med.nyu.edu Roy Smith
tpb_at_doctor.zk3.dec.com Tom Blinn
Sam.Nelson_at_cs.stir.ac.uk Sam Nelson
Guillaume.Bourque_at_Alcan.Com Guillaume Bourque
magill_at_isc.upenn.edu William Magill
anthony.miller_at_vf.vodafone.co.uk Anthony Miller
asakelaris_at_cosmote.gr Alexander Skellaris
ram_at_ini.dec.com Dr. Ram Rao
----------------------------------------------------------------------
[History]
I still haven't sold the idea of NTP to upper management yet. I am in
the process of making my pitch now. I think they will eventually buy in
but I won't synchronize without clearance. Its just the way we do
things here. Its the nature of state government. Hehehehe, the wheels
of the bureaucracy, there are many, but its the slow ones that will
crush you.
We are predominantly a Unisys 2200/532 mainframe shop, however last year
we purchased three departmental class Alpha boxes running Digital UNIX.
We also inherited seven windows NT servers from the client server unit.
Plus we are purchasing a new type of heterogeneous UNISYS/INTEL
architectured mainframe technology that UNISYS calls "ClearPath". Their
motto for ClearPath is, "While you slept the world changed".
Hehehehe, well we were sleeping, and it did change, but we're awake now
and we're running three shifts too. I think were catching up! Our
problem now is, we use to have one system clock, the mainframe's clock,
and we set it according to somebody's watch and that made us happy
because our world ran on the mainframe's time. Now we have ten system
clocks and we are buying five more and no matter how hard we try, we
can't keep all the system clocks set to the same time by hand. Each
system clock is a server, and each server has a database, and each
database has a log file, and each log file gets date and time stamped
frequently, so we came to the conclusion synchronized time is good thing
and it just might be important too! :)
Thinking this, we realized we needed to remedy the situation fast so we
began to look for solutions. We are now scrutinizing NTP because it
looks like a winner. I think we can get most of our clocks on UTC time
with manual adjustments for the mainframe as needed. To date we still
don't know if Unisys's new ClearPath technology provides support for
NTP, but we intend to find out.
Ohhh, and politically, at first, we were afraid to let an external
organization set our system clocks. With NTP, we realized, if you wanted
to, you could still keep time off someone's wrist watch and everyone could
still be happy. After checking into it deeper though, we realized we
could trust (3) or more big institutions to keep time for us, but if we
couldn't do that we could purchase a stratum one time server for less than
$5,000.00. So, to make a long story short, and its to late for that now,
the bottom line is, NTP is a good thing; if your not using it you should
be.
Reference:
http://www.ntp.org
----------------------------------------------------------------------
[Summary] What impact might NTP have on Oracle8 databases? 03/23/1999
---
Yes we run this everywhere. I am not aware of any problems it causes
(that's not to say there aren't any). Logic would suggest that you
will have more consistent data (sync events across servers) with it
on than off.
- Anthony Miller
---
---
We are using xntp ver 3 without any problems for more than half a
year. We run oracle7.3.3 on digital unix 4.0b and 4.0e, and oracle8
on nt4.sp4. Oracle seems to be happy about it, although time
adjustments if any, are in the range of microseconds.
-Alexander Sakellaris
---
---
I use ntp on dec on a cluster environment and oracle is running on top
of it, everything seems perfect for the last 2-3 years ! :)
This is exactly what ntp is all about! First it will only synchronize
back few mili second each time it synchronizes so oracle will not see
anything!
- Guillaume Bourgue
---
---
I have the setup you described running here in my office in
Indianapolis. It all works -- magic! The key is that UNIX and NTP
keep time in absolute terms (not with respect to a time-zone). It is
the responsibility of the display routines to convert time to a
time-zone relative format. hence it works out of the box with no
tweaks. Give me a call if I can clarify.
- Dr. Ram Rao
---
---
Thanks Dr. Rao, I will call, and I do frequently., Hehehehe, Looks
like your trying to answer two of my queries with one response :)
---
---
One of the fundamental principles of NTP is that time increases
monotonically. Except for a possible timewarp at boot-time (which of
course is before your Oracle instances starts) NTP doesn't jump your
clock, it slews it by temporarily changing the system's notion of the
length of each second. Thus, as far as running processes are
concerned, the seconds tick by just as they did before.
- Sam Nelson
---
---
Thanks Sam you're my hero. Seriously, you are! :)
---
---
Under normal circumstances, XNTP will NEVER set the clock backward,
unless you have it already set seriously far into the future. In that
case, you already have major problems to address. What XNTP will do
for you that's beneficial is help assure that all of the systems in
your local network share a common belief about what time it is. They
could all be wrong, but with a good source of time information
(including a remote server if you have good connectivity) they will
usually all be really close to the truth. When their local idea of
the time disagrees with the network's view (by a modest amount), they
adjust their belief a bit at a time, much more gently than anything
you can do using manual procedures
- Dr. Tom Blinn
---
---
We use NTP on all our important servers. In a nutshell, there's not
much reason to use anything else. In particular, the worry about
setting the clock backwards is a non-issue, as NTP can be configured
to ensure that this never happens. The general idea is to set the
system clock once, when you boot, and then let NTP fine-tune the clock
*rate* to keep it accurate without any backwards jumps.
- Roy Smith
---
---
Synchronizing time is something you must do if you have multiple
databases on separate machines.
- William Magill
---
---
Eeeeeeek!, O.K., O.K. were working on it :)
Hmmmmm, upenn.edu, hey, aren't you a stratum one time server? We are
looking for a 3rd server. Could we use your time services? We asked
Penn State but they haven't responded to our request yet.
---
Sincerely
Kevin Criss
----------------------------------------------------------------------
[Q] What impact might NTP have on Oracle8 databases? 03/23/1999
Reference: RFC 1305
ftp://ftp.funet.fi/pub/doc/rfc/rfc1305.txt
RFC 1361
http://www.kaska.demon.co.uk/rfc1361.htm
PUBLIC NTP TIME SERVERS
http://www.eecis.udel.edu/~mills/ntp/servers.htm
NTP - FAQ
http://www.eecis.udel.edu/~ntp/database/FAQ/faq3C.htm
There is a lot of research out there but sooooo little time. :)
We are thinking about using NTP, Network time protocol to adjust our
system clocks. We would use XNTP to synchronize the clocks of (3)
UNIX servers and (7) NT servers to network time. We are in the
process of selecting time servers and requesting permission to poll
those services.
We also have Oracle8 databases, several in fact. Before we implement
NTP we wonder what impact synchronizing time would have on living
database. Are there any gotchas we should be concerned about? Are
there any horror stories to tell? Does anyone out there use it, or
does everyone use it? What impact would setting a clock backwards
have on a running database?
Sincerely
Kevin Criss
Received on Tue Mar 30 1999 - 21:25:34 NZST