[Samba] Re: Samba 2.2.3 Oplock problem...

[Martyn Ranyard]

Remember that this is from memory of a user-group meeting about 3-4 months 
ago, but anyway...

At 04:14 PM 2/6/02 +0000, Noel Kelly wrote:
>So, Martyn would it be true to say this:
>
>Oplocks should only really be used in situations where a server is sharing
>program files or other relatively static data ?

The reason M$ created oplocks was to give an overall impression of more 
speed.  This works quite well when there is a nice tight network with no 
delay points or when it is likely that only one user is accessing the file 
at once.

>Home drives and other files which are primarily only opened by one user at a
>time would see no value in oplocks ?

No - they would see the most value, as the file is cached at the client and 
any writes remain on the client until flushed back to the server, thus 
making write access much quicker.

>Databases and other dynamic data shares which can be altered by many users
>should certainly have oplocks disabled ?

Whilst I cannot say one way or another on a tight network, because in 
theory it can slightly speed things up if say the program locks a record, 
and unlocks it, that kind of thing is never passed to the server if no 
changes are actually made.  On a leaky network you are definitely right and 
when especially when M$ Access is used, because M$ wrote it so that it 
would use oplocks heavily and that just shows up the bad behavior of such 
predictive coding.


>Thanks for your insight.
>Noel
>
>-----Original Message-----
>From: Martyn Ranyard [mailto:ranyardm at lineone.net]
>Sent: 06 February 2002 13:02
>To: Russell Senior; jra at samba.org
>Cc: Samba Maillist (E-mail)
>Subject: Re: [Samba] Re: Samba 2.2.3 Oplock problem...
>
>
>
>I remember a talk at our lug about oplocks,  If I remember correctly then
>the following is true :
>
>A client takes the file, caches it at client side, and oplocks it.
>The server then gets a different request to read the file, asks the client
>to send it's latest version.
>Here, two things can happen, 1. the client responds with changes, which the
>server reflects or 2. the client doesn't respond in time, in which case the
>server breaks the oplock and reverts the file to it's unchanged
>state.  This is the way SMB oplocks work AFAIK
>
>Network problems can cause delays and therefore timeouts will timeout (it
>is their job, afterall).  This is why leaky networks cause oplock problems.
>
>I contribute this hazy knowledge to the public domain, mainly to save
>Jeremy some time!
>
>At 08:59 AM 2/5/02 -0800, Russell Senior wrote:
> > >>>>> "Jeremy" == Jeremy Allison <jra at samba.org> writes:
> >
> >Jeremy> Most oplock problems are due to bad network setups (client
> >Jeremy> drivers, hubs etc). I haven't seen any evidence other than one
> >Jeremy> person having oplock problems (which is not unusual given the
> >Jeremy> state of many networks) that this is anything other than the
> >Jeremy> usual network related oplock woes.
> >
> >Can you elaborate on this?  As I understand it, the oplock break
> >messages are getting lost, but aren't they sent over the TCP socket?
> >Won't the regular TCP reliability guarantees ensure it gets resent if
> >not ACK'd?  How can network problems interfere?
> >
> >
> >--
> >Russell Senior         ``The two chiefs turned to each other.
> >seniorr at aracnet.com      Bellison uncorked a flood of horrible
> >                          profanity, which, translated meant, `This is
> >                          extremely unusual.' ''
> >
> >--
> >To unsubscribe from this list go to the following URL and read the
> >instructions:  http://lists.samba.org/mailman/listinfo/samba
>
>==============
>Martyn Ranyard
>
>
>--
>To unsubscribe from this list go to the following URL and read the
>instructions:  http://lists.samba.org/mailman/listinfo/samba
>
>--
>To unsubscribe from this list go to the following URL and read the
>instructions:  http://lists.samba.org/mailman/listinfo/samba

Martyn

Life's a bitch, but so am I.