[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference turris::digital_unix

Title:DIGITAL UNIX(FORMERLY KNOWN AS DEC OSF/1)
Notice:Welcome to the Digital UNIX Conference
Moderator:SMURF::DENHAM
Created:Thu Mar 16 1995
Last Modified:Fri Jun 06 1997
Last Successful Update:Fri Jun 06 1997
Number of topics:10068
Total number of notes:35879

9199.0. "Questions on new-wire-method" by NETRIX::"[email protected]" (Feynman Lo) Mon Mar 17 1997 04:27

Our most  critical local customer is running SyBase 11 on
Digital Unix 4.0b. He experienced some file corruption symptom
recently.

We have checked this notes conference and find that topic 8630
has mentioned similar problem. One of the replies suggested
to set vm: new-wire-method=0 to try to resolve the problem.

But the customer has raised questions regarding this variable.

1) What is the effect of setting this parameter to 0?
   
   Because the default is new-wire-method=1, what is the
   impact to the OS level, and applications level,  
   when re-setting this parameter to 0?

   Where can we find  the detail description of this 
   new-wire-method parameter?


2) Is it setting this new-wire-method=0 will solve the compatibility
   problem between SyBase and Digital Unix 4.0x?

   By the way, topic 8630 has not mentioned the result.


Regards,
 - feynman


[Posted by WWW Notes gateway]
T.RTitleUserPersonal
Name		DateLines
9199.1SMURF::DENHAMDigital UNIX KernelMon Mar 17 1997 09:1426
    First off, none of these (8630, 8282, 7828) has pinned file corruption
    on lightweight wiring (the thing new-wire-method enables). The problem
    there was lww casuing the system to crash. That may have well
    resulted in database corruption, but it's not exactly subtle
    corruption -- you'll know if you've got the problem.
    
    Far as I can see, no one has identified the cause of the corruption
    beyond the effects of a system crash.
    
    FWIW, lightweight wiring is just that -- a faster way to wire (lock)
    down kernel memory for I/O operations. When it works, it uses the
    PTEs of resident pages to mark the wiring rather than going through
    the more heavyweight vm-map level wiring mechanism. The system
    couldn't care less as whether you've got new-wire-method turned on
    or off. There's a slight performance penalty to turning it off, but
    that's better than a crash. The problems these notes are dealing
    with involves flaws miscommunications about the wired state of
    pages between the PTE-level wirings established by lww and the
    map level wirings. In particular, when an attempt to lw-wire a
    memory range fails because a nonresident page is encountered, the
    higher level wiring code (vm_mape_pageable) isn't being properly
    notified of the pages wired so far. This causes inconsistencies
    in the wiring state of the memory range with subsequent panics.
    
    Maybe a vm person can speak to whether similar lww confusion could
    result in corruption rather than panics.
9199.2check value of ssm-thresholdSEAWLF::COLEDigital NSIS, Greenbelt, MarylandMon Mar 17 1997 09:3314
	RE Sybase 11 on DUnix 4.0 -
	check the following:


	ssm-threshold=0      in sysconfigtab


	Helped one of my customers with similiar problems.


	larry cole
9199.3NETRIX::"[email protected]"Shashi MangalatWed Mar 19 1997 23:1112
We have seen other reports on database corruptions, which seems to have cured
by turning off new-wire-method.  As far as I know there are no CLDs or QARs
filed against the corruption.  You may want to consider filing a CLD/IPMT.

As for the LW-wire panics, a fix is in the works.  The same bug maybe
responsible for the corruption, provided the system is low in memory to cause
page stealing.  If you can consistently reproduce the corruption, I can give
you a patch to verify.  Send me a note if you are interested.

--shashi

[Posted by WWW Notes gateway]