lotto@wjh12.harvard.edu (Jerry Lotto) (09/05/88)
Wizards... please read and comment! Any useful suggestions are
deperately needed.
The following problem has been encountered on a production 3Com LAN.
As of now, there is no known solution to this. Phone calls have been
placed to the companies involved... turnaround time for this route
promises to be long and cost high.
Problem: server disk gets trashed periodically. Symptoms: when the
LAN is under heavy disk I/O load (heavy defined as one person doing
a large copy operation!), data from some random packet gets written
on top of the directory and/or FAT of the servers disk, with predictable
results.
Server Hardware: PC/AT 10MHz 0 ws clone (Microsmart) with 3C501, OMTI
ESDI controller, HP 311Mb drive formatted into 3 partitions. Partition
1 (C:) is DOS normal as required by 3+ 1.2.1 (server version) (512
byte sectors, 4 per cluster) at 32 Mb. Partition 2 has double sized
sectors for 66 Mb total space. The last partition has 2K sectors at
4/cluster for 200+ Mb (the rest of the drive). 1 Mb AST RamPage EMS
board is used by 3+ for disk buffering (see below). Recent Phoenix
BIOS.
Configuration: The disk was formatted with Vfeature deluxe (most
recent version). There are three network drives corresponding to
directories on each of the logical partitions. Everyone links to
these public directories, noone uses their own home directory. The 3+
has "turboshare" with 64 buffers/network drive in EMS memory. The
problem occurs at different combinations of buffer counts. MSDOS
version 3.1 with 20 buffers is running below the server. The only TSR
is the EMS driver.
Speculation: 3 way interaction between 3+ disk buffering in EMS
memory, DOS buffering and application software buffering. Application
swre is all tried and true (WP network version, SAS/PC, etc). Various
programs cause the problem.
Other Info: Vfeature has NOT been tested with EMS version of 3+ disk
buffering (turboshare) according to Vfeature tech support. The "real"
net guru there is on Federal jury duty for MONTHS! They will call back
(sure). 3Com talks to dealers only unless you pay them $45/hour to fix
their bugs. Dealers just recommend buying more hardware. This is a
production network... down for the count. The server cannot fit enough
buffers into non-EMS memory to boot (a diagnostic that would have been
very useful). This same hardware was running under 3+ 1.0 with an
ST506 Fuji 80 Mb drive configured the same way with vfeature (all
partitions were DOS normal, 30-30-10) with no problems for 2 years.
The previous setup used NO EMS, both 3+ itself and the individual
buffer size (cluster) were smaller so that the whole thing fit into
640K.
Questions: Does anyone familiar with some or all of these software
components know how to disable caching? Other workarounds/diagnostic
measures? Known problems with any of the components?
Thanks for your patience if you stuck with this article to the end.
Reply by mail, I will summarize all useful data to the net when
the problem is resolved.
--
Gerald Lotto - Harvard Chemistry Dept.
UUCP: {seismo,harpo,ihnp4,linus,allegra,ut-sally}!harvard!lotto
ARPA: lotto@harvard.harvard.edu