Mladen,
Do you have a recommendation regarding the use of semaphores?
I've read about the position of RH (RedHat) and LH (Left Hand storage) for
mapping out the semaphore alphabet. Flapping my arms about with red and
yellow colored flags is having no effect and I feel like a real concurrency
killer.
I don't see how increasing the number of semaphores is going to help but as
the flags wear the number of threads on them is becoming larger.
Happy 1 April.
Paul
On Thu, Apr 1, 2021, 00:10 Mladen Gogala <gogala.mladen@xxxxxxxxx> wrote:
ORA-7445 is an Oracle bug, by definition. You should open a service
request with Oracle Support. Other than that, shared memory is an operating
system trick. Virtual memory is based on page tables. There is a table
somewhere in the OS which contains virtual addresses and their
translations. Translation can be an address of a physical page in RAM or
can be a page in a page file. If two processes map the same page table
into their address space, we talk of shared memory. Yes, there are some
details about that, such mapping is performed by shmat system call and not
by sbrk, malloc or calloc. Shared memory is used for SGA. You can see the
shared memory on your system by using ipcs -m.
PGA, on the other hand, is allocated from the process private memory. That
means that PGA consists of segments which are only mapped to a single
process. PGA is allocated using sbrk, malloc or calloc. If PGA is expanded
beyond limits, it will eventually hit another segment or OS limit and
produce SIGSEGV "segmentation violation" signal and produce ORA-04036. Note
that ORA-4036 is NOT the same as ORA-7445. ORA-7445 is a bug which needs to
be resolved by Oracle Support. If you want to learn the basics of the Linux
operating system, I would recommend 2 books:
1. Understanding the Linux Kernel by Daniel Bovet
2. Modern Operating Systems by Andrew Tannenbaum
Both books are available on Amazon. Of course, the unspoken prerequisite
for both books is certain book by Kernighan & Ritchie which should be
taught to K12 children (ANSI edition, of course). What you are asking is
quite complex to explain. The usual answer is that PGA and SGA are "two
different types of memory" but I don't find that answer satisfactory. In
order to understand the memory handling on Linux, you would need to
understand interrupts, paging, processes, page tables, address spaces and
IPC (Inter-Process Communication). PGA is allocated from the process
private address space, SGA is allocated from shared memory. Now what is
private memory space, how does Linux manage it, what is a page fault and
other questions are rather difficult to explain. Of course, you are the
only person who can decide whether you need deep knowledge of the Linux or
not.
On 3/31/21 3:03 PM, Amit Saroha wrote:
Dear All,
Please help me understanding PGA and SGA.
I was under the impression that PGA and Memory and SGA Memory are
differently allocated. But, in a recent situation, we observed *ORA-7445
[kghalf] *and my DBA is saying - out of a total of 256 GB system memory
one session's PGA consumed everything and only 8 MB was left for other
sessions. But, apart from one process, no other processes were impacted so
I have doubt that - Oracle consumes SGA memory and Allocates it to PGA of a
single process.
Please help me understand.
--
Mladen Gogala
Database Consultant
Tel: (347) 321-1217https://dbwhisperer.wordpress.com
-- //www.freelists.org/webpage/oracle-l
