Article 39025 of comp.databases.oracle:
Newsgroups: comp.databases.oracle,de.comp.databases,comp.os.linux.misc
Path: psuvax1!uwm.edu!news.moneng.mei.com!sol.ctr.columbia.edu!xlink.net!news.ppp.de!news.Hanse.DE!mwhh!deerwood!georg
From: georg@deerwood.hanse.de (georg)
Subject: Re: LINUX featuring ORACLE 7
Followup-To: comp.databases.oracle,de.comp.databases,comp.os.linux.misc
X-Newsreader: TIN [version 1.2 PL2]
Organization: Private Site
Message-ID: <DArG43.9wE@deerwood.hanse.de>
References: <1995Jun6.010843.12307@deerwood.hanse.de> <D9wHwq.MBG@nl.oracle.com> <1995Jun17.175442.7681@deerwood.hanse.de>
Date: Mon, 26 Jun 1995 03:24:02 GMT
Lines: 105
Xref: psuvax1 comp.databases.oracle:39025 comp.os.linux.misc:60359

Hello netters,

some more results of my testing of SCO Oracle running on a Linux Box
using the Linux iBCS2 emulator:

+++ performance

I did a little performance test. It is written in oraperl (an
Interactive Unix version of oraperl, linked with shared libraries, again
using the iBCS2 emulator). The test inserts 1000 rows into a simple
table, then commits. The table has 5 numeric columns, one of which is
declared as the primary key and thus has an implicit index. The values
for this column are taken from a sequence, the values  for the other
columns are assigned at runtime. One test is done by over and over
opening the modified statement again, the other does it by binding the
new values to the statement.

The 2 tests were done 100 times, so after that I had 200,000 rows
inserted. The used hardware is:

486 DX2 66 MHz PCI
16 MB main memory
NCR 53c810 SCSI controller
Conner CFP-1060S hard disk

Software configuration was:

mix between SMALL and MEDIUM config in init.ora:
db_file_multiblock_read_count = 16 # MEDIUM
db_block_buffers = 550             # MEDIUM
processes = 100                    # MEDIUM
log_buffer = 32768                 # MEDIUM
timed_statistics = true # because I do tracing often
all other parameters default or set to SMALL configuration.
Rule based optimizer used.

The startup times of oraperl are included in the test time. The results
(tps is transactions per second):

                begin   end     min     avg     max

test1, reparse of statement
  real:         41.0    33.0    31.0    33.0    43.0
  user:          3.5     3.3     2.9     3.3     3.9
  system:        2.4     2.3     1.8     2.2     2.8
  ORA-user:     19.4    20.0    19.2    19.9    20.7
  ORA-system:    3.3     2.8     2.5     3.0     3.7

  tps:          24.4    30.3    32.3    30.3    23.3

test2, one parse, using 'bind' mechanism
  real:         23.0    16.0    16.0    17.9    23.0
  user:          2.3     2.5     2.0     2.5     2.9
  system:        2.6     2.1     1.9     2.3     2.6
  ORA-user:      6.9     6.8     6.2     6.8     7.3
  ORA-system:    2.7     2.4     2.1     2.6     3.0

  tps:          43.5    62.5    62.5    55.9    43.5

The Linux system had nothing else to do, but X was started with several
xterms, a total of 60 processes was running at test time. There was no
performance loss with increasing table size. Instead in the beginning
the higher real times are from swapping. The times stabilize after doing
about 3 rounds and remain stable over the other 97 rounds.

And an additional note: I just did a similar test on our HP9000-800, a
system selled for database activities. The machine was idle, as the
Linux box was. The inserts using the 'bind' have real execution times
from 6 to 9 seconds giving a tps of 166 to 111. Thus it 'only' is about
3 times faster.

+++ access to database

As you may expect from the above, it is possible to access the Oracle
database using 'oraperl' (Authors: Larry Wall / Kevin Stock).

All Oracle tools (SQL*Plus, SQL*DBA, exp, imp, sqlldr) work.

Additionally I tested access from an OS/2 and Windows client via SQL*Net
TCP/IP V1. On the PC side I used Personal Oracle V7.1.4.1.0. I used
SQL*Plus for Windows (directly using SQL*Net DLL's) and a Visual Basic
application, that accesses the database via ODBC. Both work fine.

A third access method was tested with a self-written networking component.
On the client side it does a Winsocket connect to a special port on the
Linux system. On the Linux side 'inetd' starts a self-written server
(compiled under Interactive Unix) containing embedded SQL code to access
the database. Works.

+++ conclusion

SCO Oracle runs absolutely stable (for 3 weeks) and reasonable fast on
Linux. The greatest pitty is, now I know, all works fine, I have to
remove my installation, because of the missing license -- but I still
have some days evaluation left, I hope.

I recommend: request Oracle Versions for Linux from Oracle as often as
possible, may be, they will listen to customers?

regards
Georg
-- 
 ___   ___
| + | |__ ' Georg Rehfeld,  D-20535 Hamburg, Jordanstr. 8,  [49] (40) 2518356
|_|_\ |___, georg@deerwood.hanse.de