IPC Shared Memory sub-system on IBM AIX 3.1.5

IPC Shared Memory sub-system on IBM AIX 3.1.5

Post by ED ZIETARS » Fri, 23 Oct 1992 22:59:01



I am porting some code from a Siemens Nixdorf WX 200 UNIX system to IBM RS/6000
AIX 3.1.5 (3005). The code makes use of the IPC Shared Memory sub-system. It is
claimed that the existing code has UNIX system calls compliant with "The X/Open
Portability Guide III" (or XPG3).

Are there any particular differences/limits/limitations/bugs in the IPC Shared
Memory sub-system implementation on AIX 3.1.5 (3005) ?

Thanks in advance for any help.

(email is preferred and I will post a summary to these newsgroups.)
--
Ed Zietarski, Boldon James Limited, Damian House, West Street, Congleton,
CHESHIRE CW12 1JN, UK (Snail), +44 260 280465 (Phone), +44 260 298030 (Fax),

..!mcsun!ukc!pyrltd!bjcong!emz (Bangnet).

 
 
 

IPC Shared Memory sub-system on IBM AIX 3.1.5

Post by ED ZIETARS » Thu, 29 Oct 1992 02:36:54


The AIX manual page for the IPC Shared Memory function shmget() states that the
following limits apply:

        Maximum shared memory segment size is 228 bytes.
        Minimum shared memory segment size is 1 byte.
        Maximum shared memory memory IDs is 4096.

Are these limits different to those found under other implementations of UNIX ?

I need to port some code which uses the IPC Shared Memory interface. The
software was originally developed on a Nixdorf Siemens WX 200 UNIX system.

It is claimed that the system calls are compliant with the X/Open Portability
Guide III (XPG3).

On Pyramid under the AT&T universe, the same manual page does not list any
limits.

Any comments ?
--
Ed Zietarski, Boldon James Limited, Damian House, West Street, Congleton,
CHESHIRE CW12 1JN, UK (Snail), +44 260 280465 (Phone), +44 260 298030 (Fax),

..!mcsun!ukc!pyrltd!bjcong!emz (Bangnet).

 
 
 

IPC Shared Memory sub-system on IBM AIX 3.1.5

Post by sc.. » Fri, 30 Oct 1992 02:59:52



> The AIX manual page for the IPC Shared Memory function shmget() states that the
> following limits apply:

>    Maximum shared memory segment size is 228 bytes.
>    Minimum shared memory segment size is 1 byte.
>    Maximum shared memory memory IDs is 4096.

It appears you were running the ASCII version of info which has a problem
showing the '^' character.  The maximum shared segment size is 2^28 or 256MB.
 
 
 

1. Oracle 8.0.5, aix 4.3.2 & VSS Raid Sub-System

Hi all,

Can anyone out there offer advice on how to configure an IBM VSS disk
subsystem to give reasonable performance with an Oracle 8.0.5 DB.
Current throughput rates are around 9Mbs for read and only 1.5Mb for
read/write

The application is a DSS and is of the order of 150Gb.

The configuration is an IBM VSS with 8 drawers, each drawer has two RAID
5 arrays on the same SSA loop with each array being divided up into 2 x

I don't want any replies stating the obvious, that you shouldn't use
RAID 5 etc. for DSS systems, the point is we've been lumbered with one
and we have to get the best from it.

IBM recommend using 32K striping and using as many SCSI adapters as
possible. However there seems to be a real problem with parallel queries
generating what seems to, the subsystem & it's cache, random I/O when in
fact it's really sequential table scans and therefor fails to provoke
the read cache into performing read-ahead.

We've actually tried not striping the data and improved read-only
queries by trying to predict where the Oracle optimiser will start each
parallel query and laying out the raw LVs used by the tables to
coincide, but it's a lot of work.

Read/Write performance is especially low at perhaps 1/6th of the
read-only rate. This makes me think that the write cache is not being
invoked in "lazy" write mode or something similar although even then I'd
only have expected the rate to quarter. However, perhaps two-phase
commitment is accounting for the remainder.

So my questions are :-

What rate of throughput should I expect from my SCSI adapters when I'm
doing read-only & 50% read/write workloads.

What, if anything, can I do about the disk layout which will improve
performance

What, if anything, can I suggest to our Oracle programmers to bend the
application to get more from the hardware we're stuck with ?

I'd also welcome views on, if I reverted to classical SSA disks, how
should I best configure them for Oracle Parallel Queries, JBOD, RAID,
should I stripe, if so at what level 4k, 8k, 32K or at the LVM partition
level.

On a related matter, I personally think using LVM stripping isn't worth
the bother. I don't see the performance gain outweighing the loss of
mirroring type facilities for online backups, migration etc. and it
makes disk replacement a nightmare. Comments please.

TIA

John McQue

IMS Health

2. Passwords

3. AIX sub-system refresh problem

4. Compile error in 2.5.38-mm1

5. Sub-sub-sub-sub-sub subdomains?

6. Problem; setting DHCP slows down everything

7. System V, IPC Shared Memory

8. unpv12e source code

9. Fortran & IPC using System V shared memory

10. Allocating shared memory IBM S80 / AIX 4.3.3 / IDS 7.3

11. AIX Shared Memory Segment (Not enough memory)

12. Sub-system problem