SUMMARY: Non-deterministic behaviour on Ultra SPARC

Hi,

I got only one reply (which asked if e.g. /var was full, which was not the case)
to my question (the original question is attached below).

In the end with the help of our hardware dealer I succeeded to contact someone at
SUN.
He advised to apply some patches:

106625-11
106292-11
105786-14
105516-06
105529-11       
105580-17
105379-06
106439-07
106842-09
106841-01

After that the problem vanished :-)
Although some of the above patches obviously are not related to our problem
(e.g. the pkgadd patch) at least one seemed to correct a real problem.
My personal favorite is the glm driver patch (105580-17).

Cheers,

Andreas

--------------------------------------------------------------------------------
> SunOS cavallo 5.6 Generic_105181-26 sun4u sparc SUNW,UltraSPARC-IIi-Engine
> 
> which shows a reproducible non-deterministic behaviour.
> We first observed it as "corrupted blocks" in an oracle database and nailed
> it down to the following:
> Under load simple reading of big files (i.e. 100 MB and larger) gives
> diefferent results each time the file is read.
> If the load goes down - reading is OK.
> (We tested it with many parrallel md5sum's which pushed the load above 10).
> No entries in syslog or dmesg :-)
> The behaviour is not bound to a specific disk or controller.
> 
> prtdiag -v says:
> 
> System Configuration:  Sun Microsystems  sun4u SPARCengine(tm)Ultra(tm) AXi (UltraSPARC-IIi 360MHz)
> System clock frequency: 90 MHz
> Memory size: 1024 Megabytes
> 
> ========================= CPUs =========================
> 
>                     Run   Ecache   CPU    CPU
> Brd  CPU   Module   MHz     MB    Impl.   Mask
> ---  ---  -------  -----  ------  ------  ----
>  0     0     0      360     0.2   12       9.1
> 
> ========================= IO Cards =========================
> 
> No failures found in System
> ===========================
> 
> ========================= HW Revisions =========================
> 
> ASIC Revisions:
> ---------------
> 
> System PROM revisions:
> ----------------------
> 
> Anybody with any ideas? Do we have a defect CPU or Cache,
> 
> TIA
> 
> Andreas