ntp server

[herzel]

Hi All
I am UNIX/Linux sysdmin in Telecom Company. There was request in one of our branch companies to implement ntp and synchronize times exactly at the millisecond. I noticed that there are differences of 0.1 - 0.2 seconds between the servers after the ntp implementation took care of.
Is it sense able?
Does this amount can be zero I mean without any difference at all?
If I cannot force the clients to be exactly the same time can I make change that one client be synched with ntp always but with configured value of time behind the other servers?

Thanks in advance
(I know this forum is the best one)

 

[thedaver]

I would recommend that you challenge the "to the millisecond" requirement. It is BARELY feasible to use NTP to synchronize two ntp clients to the millisecond.

THEN, due to drift of internal clocks/systems, you WILL have almost immediate differences between any two machines taking time source from the same NTP server.

D.E.R. Management - IT Project Management Consulting

http://www.dermanagement.com/

 

[macd68]

Not a valid requirement in my opinion. I've never had or heard of an operational mandate for millisecond precision on many disparate hosts utilizing ntp. Some greater granularity is to be expected.

 

[zeland]

<sarcasm level="high">
herzel: I've found the solution to our time accuracy problems down to the micro second. You just need to approve these forms.
Manager: What are these for?
herzel: The first set of forms are to build an underground data center with 9 inch lead walls and state of the art biometrics security.
Manager: What in blazes do we need that for? It sounds like a nuclear bomb shelter!
herzel: You are most correct. As you can see in the second set of forms, those are the specs given by the Department of Environment.
Manager: Why are they involved in all of this?
herzel: Well, if you have noticed the third set of forms, the purchase of our new servers with high precision atomic clocks & thermo nuclear power supplies are highly regulated.
Manager: And this will set us back by how much exactly?
herzel: I don't have the final figures yet, as human resources is still finalizing the hiring of a hazmat team.
Manager: Hazmat team, hmmm... Can we use some of the accountants to man this new data center?
herzel: I suppose we could, but we might need to up their insurance premiums.
Manager: Bah!
herzel: We'll be the top of our industry with this new technology. Imagine the extra revenue due to our highly accurate billing. You might be inline for a promotion too.
Manager: Higher revenue, yes. Revenue is good. Get on this straight away!
herzel: Yes sir!
</sarcasm>

--== Anything can go wrong. It's just a matter of how far wrong it will go till people think its right. ==--

 

[Knackster]

ROFL!

I swear I was reading one of Simon's BOFH articles!

GJ!

Chris
IT Manager
Houston, Texas

 

[macd68]

Good stuff.

 

[herzel]

Great
I am happy that you can laugh on me.
I hope this post will not go to the authority servers or else they will think I am developing some nuclear bomb and will arest me.
For our case: our DB department had made this request.
They compile on sun machine get the DB files from netapp storage and The DB server is at all Linux machine (as you may be familiar the db is IBM ClearCase). They got at the end of the compilation an error that discovered to be because of times are not synced between the 3 machines.
I did implement ntp hoping to sync the machines. But the DB people still complaining.
So I asked the best forum I know.And for some unknown reason zeland makes fun of me!!!!!

 

[thedaver]

herzel; I'm sure the joke was at the expense of a very silly functional requirement and wasn't intended as personal...

As to reaching a solution, perhaps you should elaborate a bit more on the root problem that the DB team is trying to solve. It doesn't sound very wise to have a time-based dependency between several machines... noting the thread's progress already.

If you wanted to share a bit more of the details, I'm confident that this forum's talent would take up a challenge in a very positive manner..

Otherwise "sync time to the millisecond" has already earned its due.

D.E.R. Management - IT Project Management Consulting

http://www.dermanagement.com/

 

[chipperMDW]

And for some unknown reason zeland makes fun of me!!!!!

It looked like he was making fun of "Manager" for having made a request of you that is unreasonable and probably doesn't even address the real problem at hand. Compare the tone to that in nearly any installment of the Dilbert comic strip.

 

[zeland]

herzel;
I'm sorry if you took offense to the BOFH/Dilbert style joke I posted. Thought that I had the post sufficiently marked/labeled as such. I've always noticed that when people are under pressure and stress, their ability to solve the simplest of problems will be at an all time low. I like to send these type of messages to my people to make them forget about their immediate problems for just that one moment and make them think "out of the box" with twisted humor & sarcasm.

As thedaver said, it is almost impossible to avoid any type of time skewing on any machine. As you should know, the timing component in the RTC of computers are made from quartz crystals. Depending on the quality of the crystal and voltage applied, the resonance of the crystal will also vary effecting its accuracy. That was why I brought up the joke about computers with atomic clocks. So unless your computers have atomic clocks in them, you can't possibly expect that kind of time resolution. I experimented with some of my computers and found out that the cheap motherboards could give me time skews of about +/-0.6 seconds. The branded and more expensive ones would have about +/-0.002 to +/-0.005 seconds skew.

My brief read up about ClearCase indicates to me that it is a software/source code version control system. From my understanding, VCSs can tolerate a few seconds of time skew when files are checked in or out. You may want to ask your DB administrators what is the toloreance of CleareCase and if it can be adjusted. Another thing that I briefly skimed through is that one of your VCS servers may have had a "very large" time skew during one of the checkins and ClearCase is having problems consolidating the sources after its time has been readjusted. You may want to check this by checking out some older revisions.

--== Anything can go wrong. It's just a matter of how far wrong it will go till people think its right. ==--

 

[Al42]

A solution to the problem, BTW, is GPS - have a GPS-driven hardware clock running at each computer, and have the system clock sync to the hardware clock every minute or so. You can achieve microsecond accuracy that way.

 

[sillyVM]

if you have large number of servers across multiple networks, it will be hard to be milliseconds accurate. some time hardware clock can be extremely accurate, but sync with your servers will be a big problem.