Here is the environment for my test
1. I did test over both network and localhost. The result is the same.
2. I also did 4 machine test, which machine run 4 instance.
3. The test is basicly a simple insertion. Did the insertion through statement and perpare statement. The data is no more than 65 bytes insert into a single table. So there may have some table lock issue.
4. I set the MAXCPU to 4.
Here is what I find out after spend days on it.
the bottleneck is not CPU power. It is the disck IO. Windows NT always use buffered IO on the io system. Use can use the PerfMon to watch the disk queue length jump when the requests come in. UNIX has two kind of IO. The character device will physical write data to disk. I test the Sybase Database two. Sybase, which is the base of M$ SQL has about the same performance on my Sun box as the M$ SQL on PC Server. But Sybase can change the configuration to use buffered IO, which 2 times faster than M$ SQL.
So my question is whether SAP DB can do the same or not. If it can, how do I configure it. Although the data might not save when it is in the memory if the server crash, but performance is much better.
"Schroeder, Alexander" <[EMAIL PROTECTED]>wrote:
Hello,(1) Did you the test over the network, or local on the machine? (If you did it remotely, of course be sure that the networkis not the bottleneck of all (ie. has the PC a better network connection).(2) What did you use to insert the data. I.e. did you use prepared statement, or normal statement, or an addBatch/executeBatch?(I'm just curious).(3) If you did run the test on the Sun machine local, try to experiment with a MAXCPU=3 parameter for the test.Anyway, (of course depending on the scenario), often such 'single user tests' do not say much about a system's performanceunder load. There is always a tradeoff between being fast for 1 user and being also acceptable for 100 parallel users.IMHO, the load time does not play a big role, as this happens mostly 1 time in your application, and this can be easily organizedto be at startup time or similar, to not put this burden on the first request which accesses the database.RegardsAlexander Schr�derSAP Labs Berlin-----Original Message-----
From: XT Wang [mailto:[EMAIL PROTECTED]]
Sent: Sunday, September 29, 2002 4:16 AM
To: [EMAIL PROTECTED]
Cc: Stephen Gutknecht (SAPDB)
Subject: RE: SAP DB vs MS SQL PerformanceThanks for the repy. First I need to clearify two thing
1. The basic test is against Microsoft SQL Server not MySQL
2. We are not try to find out which one is better. We try to justify why we should move from PC Server Database to UNIX server.
The test is through the JDBC, which is our application environment.
The question is why M$ SQL (173,609ms) is twice fast than SAPDB (319,399ms) to insert 10,000 64 byte record. I am not sure JDBC drive play what kind of role this performance. One thing I notice is JDBC drive for SAPDB take 150ms than to load than Mircrosoft JDBC driver. I used default configuration for M$ SQL. The SAPDB configuration is used the one from the website. Here is the configuration parameters.
param_rmfile
param_startsession
param_init OLTP
param_put MAXCPU 4
param_put DATA_CACHE 16384
param_put MAXDATAPAGES 32768
param_put MAXUSERTASKS 100
param_checkall
param_commitsession
param_adddevspace 1 SYS SYS01 F
param_adddevspace 1 LOG LOG01 F 10000
param_adddevspace 1 DATA DATA01 F 20000
db_cold
util_connect
util_execute INIT CONFIG
util_activate DBA,dba
load_systab -ud dbaBefore I expect the SAPDB run on 4 CPU 2GB Memory 64bit Solaris should be at least twice faster than M$ SQL. There go to be something wrong. Need help to figure it out.
Thanks
"Stephen Gutknecht (SAPDB)" <[EMAIL PROTECTED]>wrote:
You are either going to get a lot of replies or very few :)
First off, lots of questions. Example: just because you have 2GB of RAM -
did you allocate it to SAPDB? What about the hard drive configurations?
MySQL is known to be very fast and very simple. www.slashdot.org has
several threads on this discussion (any database vs. MySQL).
How did the concurrent test work? Did you have 10 programs do 1,000 inserts
in a loop?
Personally, I would not expect SAPDB to be faster than mySQL... "more
powerful" != faster. I also find that the SAPDB network database drivers /
protocols are not the greatest for small and frequent transactions...
Have you tried running your test against localhost for both MySQL and SAPDB?
You might find a dramatic difference between local server + client and
over-network.
Just my obvious feedback as another user! ! ! ! . One could write books on a topic
this complicated.
Stephen
-----Original Message-----
From: XT Wang [mailto:[EMAIL PROTECTED]]
Sent: Saturday, September 28, 2002 10:29 AM
To: [EMAIL PROTECTED]
Subject: SAP DB vs MS SQL Performance
Hi,
We are moving database from MS SQL on PC server to Sun E450. The current
database is MS SQL 2000 on a Single CPU Dell 1300 PC Server (550MHz PIII
with 1GB Memory). E450 is 4 CPU (450MHz) with 2GB memory. We thought SAPDB
should be more powerful than MS SQL. We did a simple test after installed
the SAPDB64 on Solaris 2.8 (64bit). The result surprise us. We use three
clients to concurrently insert 10,000 small record (64 bytes) into db. The
average time is 37, 216ms on SAPDB on 4 CPU Sun Box. But MS SQL only used
18,764ms to do the insertion. The default setting ws used on both SAP DB and
MS SQL. This picture is deadly wrong. There should be no way a 4 CPU S! ! un ! ! Box
slower than PIII Intel box. Any suggestion what can cause it
Thanks
Dave
Do you Yahoo!?
New DSL Internet Access from SBC & Yahoo!
Do you Yahoo!?
New DSL Internet Access from SBC & Yahoo!
Do you Yahoo!?
Faith Hill - Exclusive Performances, Videos, & more
faith.yahoo.com
