RDB/VIDAO vs. SQL*Plus, network performance issue?
Date: Mon, 21 Feb 1994 19:10:25 GMT
Message-ID: <2kb0v1$22t_at_wu1.wl.aecl.ca>
Hello all,
I am a network administrator, and I am looking for someone with some database expertise to advise me on a remote database access performance concern.
Our database and VAX system administration people have been doing some tests comparing DEC's RDB/VIDAO accessworks product for Oracle to Oracle's SQL*Plus, both using Oracle SQL*Net for network access to our Oracle server.
They have been experimenting with transferring data blocks between the sites via our 56kb/s WAN link as well as over our ethernet LAN and we are currently experiencing low data transfer rates in both cases.
My question is: Is the poor performance of this database client (RDB/VIDAO) related to network speed or tuning, or is it an internal problem? I have provided several results from our testing as examples of the problem. DEC has concluded that this is an RDB performance problem and has said it will refer it to their RDB (Engineering) group for solution. We want to confirm that this is not a network performance issue. My hunch is that network performance is not a dominant issue, since test 2 shows that WAN access is 7 times slower than LAN access, whereas the actual WAN link in use is roughly 160 times slower than the LAN.
Test Configuration
uVAX-1 - uVAX 3100 model 80 located at Local Site on LAN VAX-2 - VAX6620 located at Local Site on LAN VAX-3 - VAX6620 located at Remote Site via WAN uVAX-1 and VAX-2 are connected on a cluster set, which are connected via a 64 kb/s satellite link (bridged) to VAX-3 at our remote site. VAX-2 and VAX-3 have the following process quotas for the account used in these tests: Maxjobs: 0 Fillm: 200 Bytlm: 30000 Maxacctjobs: 0 Shrfillm: 0 Pbytlm: 0 Maxdetach: 0 BIOlm: 22 JTquota: 16384 Prclm: 6 DIOlm: 22 WSdef: 1024 Prio: 4 ASTlm: 28 WSquo: 1024 Queprio: 0 TQElm: 10 WSextent: 51200
Test 1
Transfer using AccessWorks set of products - WAN access (i.e. read the table Using VIDAO product and create a copy of it in RDB on uVAX-1)
Date From To File Elapsed Cpu Buffered Direct
Node Node Size Time time IOs IOs
94/01/18 VAX-3 uVAX-1 3.3Meg 14.3 Hrs 56.3 sec 126988 2838 (WAN) 94/01/19 VAX-3 uVAX-1 3.3Meg 14.6 Hrs 58.6 sec 128782 2793 (WAN)
Test 2
Transfer the same table using Oracle/SQL*Net from VAX-3 to VAX-2 for comparison. Then transfer this table from VAX-2 to uVAX-1 through AccessWorks
(i.e. read the table on VAX-2 using VIDAO and create a copy of it in RDB on uVAX-1).
Date From To File Elapsed Cpu Buffered Direct
Node Node Size Time time IOs IOs
94/01/21 VAX-3 VAX-2 3.3Meg 0.75 Hrs 15.0 sec 4016 318 (WAN)
94/01/21 VAX-2 uVAX-1 3.3Meg 0.11 Hrs 48.0 sec 131177 2797 (LAN)
Test 3
Compare the efficiency of RDB versus ORACLE to select the same amount of data from the remote node, using the SQL SELECT command.
Date Product From To File Elapsed Cpu Buffered Direct
Node Node Size Time time IOs IOs
94/01/31 RDB VAX-3 uVAX-1 3.3Meg 6.00 Hrs 83.2 sec 134287 8761 (WAN)
94/02/01 ORACLE VAX-3 uVAX-1 3.3Meg 0.45 Hrs 28.5 sec 5087 8015 (WAN)
Test 4
Compare the efficiency of RDB versus ORACLE to select the same amount of data using SQL select between clustered nodes VAX-2 and uVAX-1.
Date Product From To File Elapsed Cpu Buffered Direct
Node Node Size Time time IOs IOs
94/02/14 ORACLE VAX-2 uVAX-1 175KB 19.9 sec 3.11 sec 524 324 (LAN)
94/02/14 RDB VAX-2 uVAX-1 175KB 228 sec 13.3 sec 6446 4054 (LAN)
Thanks in advance,
--Received on Mon Feb 21 1994 - 20:10:25 CET
-------------------------------------------------------------------------
Larry Gadallah Internet: gadallah_at_wu1.wl.aecl.ca UNIX Support, AECL Research Whiteshell Voice: (204) 753-2311 x2603
-------------------------------------------------------------------------