David Jones
June 12th 03, 08:08 PM
Does this also happen when using a mapped drive instead
of the UNC (\\server\share)? Does this test program
close the connection after each write - ie, is it doing a
connect/close/connect/close cycle 10,000 times?
Also, are there any WINS servers on the network in
question?
>-----Original Message-----
>I have found that Windows XP and 2000 workstations are
>consistently 10 times slower than equivalent Windows 98
>workstations in local database operations where the
>database is stored on a network drive.
>
>I created a test program to simulate a database
>called "DiskTest" (http://www.fhcc.com/downloads.htm).
>This program creates a 10MB test file by writing 1KB of
>data to the file 10,000 times. "DiskTest" also has an
>optional BDE test which requires that you have Borland's
>Database Engine installed.
>
>The tests were conducted by copying the program to a
test
>folder on the network drive, and running the program.
The
>test file is created in the programs default directory.
>
>I have tested this performance issue on Windows 2000
>servers and workstations, Windows XP Home and Windows XP
>Professional, Windows 98 SE, Dedicated server networks
and
>Peer-to-peer networks. The results are the same. If the
>test program is run on a Windows 2000 or XP computer,
the
>test takes approximately 20 seconds (100 MB, Full
Duplex).
>If the test is run on a Windows 98 computer, the test
>takes approximately 2 seconds.
>
>The simplest way to recreate the test is to use one
>Windows 98 computer connected to one Windows XP or 2000
>computer.
>1. From the Windows 98 computer, browse to a network
>share on the Windows XP/2000 computer.
>2. Copy DiskTest to the Windows XP/2000 share and
run
>the test. Note the speed.
>3. Move to the Windows XP/2000 computer used in step
1
>4. Browse to a network share on the Windows 98
>computer.
>5. Copy DiskTest to the Windows 98 share and run the
>test. Note the speed.
>
>You could also add a 3rd Windows 2000 or Windows XP
>computer. If you did, you would find that it did not
>matter what type of network drive you used (Windows
2000,
>XP or 98). What matters is the computers operating
system.
>
>Your results will vary with computer speed, network
speed
>etc. Since NOTHING is changed EXCEPT where the test
>program is being run from (same network hardware, same
>computers), you would expect the speed to be similar. I
>consistently is not! I have run this test on twenty
>different networks, with different hardware manufactures
>network configurations, and the result is always the
same:
>
>Windows XP/2000 is ten times slower that Windows 98!
>
>Is should also be noted that there are no delays
starting
>up, no difficulty browsing the network, no noticeable
>speed difference copy files around using Windows
Explorer!.
>
>I moved the testing back to my office where I performed
a
>network capture of the test file being created. A
summary
>of the test results is:
>
> indows 2000 Windows 98
># Frames Captured 157284 12247
># Bytes Captured 27897394 10909910
># Frames Dropped 0 0
>
>Note that the same 10,000 requests for 1K "records"
>resulted in 157,284 packets in Windows 2000/XP and
12,247
>in Windows 98 - a 10 fold increase!. Notice the increase
>in bytes as well 28 Million in Windows 2000/XP and 11
>Million in Windows 98!
>
>The capture indicates that Windows 98 used the SMB, NBT
>and TCP protocols, created the file, wrote the 10,000 1K
>records, and then closed the file, with a small amount
of
>overhead. Windows 2000/XP used the SMB Protocol
>exclusively, opened the file, send thousands or READ and
>QUERY frames, and eventually wrote the file in 100 Byte
>Chunks.
>
>I don't know if this is the way it is supposed to work,
I
>just know that this is how it IS working.
>
>Has anyone else seen this? Does any on have an
explanation
>or a solution?
>
>.
>
of the UNC (\\server\share)? Does this test program
close the connection after each write - ie, is it doing a
connect/close/connect/close cycle 10,000 times?
Also, are there any WINS servers on the network in
question?
>-----Original Message-----
>I have found that Windows XP and 2000 workstations are
>consistently 10 times slower than equivalent Windows 98
>workstations in local database operations where the
>database is stored on a network drive.
>
>I created a test program to simulate a database
>called "DiskTest" (http://www.fhcc.com/downloads.htm).
>This program creates a 10MB test file by writing 1KB of
>data to the file 10,000 times. "DiskTest" also has an
>optional BDE test which requires that you have Borland's
>Database Engine installed.
>
>The tests were conducted by copying the program to a
test
>folder on the network drive, and running the program.
The
>test file is created in the programs default directory.
>
>I have tested this performance issue on Windows 2000
>servers and workstations, Windows XP Home and Windows XP
>Professional, Windows 98 SE, Dedicated server networks
and
>Peer-to-peer networks. The results are the same. If the
>test program is run on a Windows 2000 or XP computer,
the
>test takes approximately 20 seconds (100 MB, Full
Duplex).
>If the test is run on a Windows 98 computer, the test
>takes approximately 2 seconds.
>
>The simplest way to recreate the test is to use one
>Windows 98 computer connected to one Windows XP or 2000
>computer.
>1. From the Windows 98 computer, browse to a network
>share on the Windows XP/2000 computer.
>2. Copy DiskTest to the Windows XP/2000 share and
run
>the test. Note the speed.
>3. Move to the Windows XP/2000 computer used in step
1
>4. Browse to a network share on the Windows 98
>computer.
>5. Copy DiskTest to the Windows 98 share and run the
>test. Note the speed.
>
>You could also add a 3rd Windows 2000 or Windows XP
>computer. If you did, you would find that it did not
>matter what type of network drive you used (Windows
2000,
>XP or 98). What matters is the computers operating
system.
>
>Your results will vary with computer speed, network
speed
>etc. Since NOTHING is changed EXCEPT where the test
>program is being run from (same network hardware, same
>computers), you would expect the speed to be similar. I
>consistently is not! I have run this test on twenty
>different networks, with different hardware manufactures
>network configurations, and the result is always the
same:
>
>Windows XP/2000 is ten times slower that Windows 98!
>
>Is should also be noted that there are no delays
starting
>up, no difficulty browsing the network, no noticeable
>speed difference copy files around using Windows
Explorer!.
>
>I moved the testing back to my office where I performed
a
>network capture of the test file being created. A
summary
>of the test results is:
>
> indows 2000 Windows 98
># Frames Captured 157284 12247
># Bytes Captured 27897394 10909910
># Frames Dropped 0 0
>
>Note that the same 10,000 requests for 1K "records"
>resulted in 157,284 packets in Windows 2000/XP and
12,247
>in Windows 98 - a 10 fold increase!. Notice the increase
>in bytes as well 28 Million in Windows 2000/XP and 11
>Million in Windows 98!
>
>The capture indicates that Windows 98 used the SMB, NBT
>and TCP protocols, created the file, wrote the 10,000 1K
>records, and then closed the file, with a small amount
of
>overhead. Windows 2000/XP used the SMB Protocol
>exclusively, opened the file, send thousands or READ and
>QUERY frames, and eventually wrote the file in 100 Byte
>Chunks.
>
>I don't know if this is the way it is supposed to work,
I
>just know that this is how it IS working.
>
>Has anyone else seen this? Does any on have an
explanation
>or a solution?
>
>.
>