Did you have the chance to run the test locally on the AR System server ? With the user tool if it's on Windows or runmacro/driver if you are on Linux ?
Or maybe a simple file tranfer of a few megabytes from your MT (in the arsys/shared folder for example) to your workstation ?
You can also compare the TCP streams with a tool like Wireshark.
Yes I did.
I've connected to both application servers using RDP, then used the user tool installed on the applicaton server itself to perform the test (and also perform the same test from within SQL developer).
I have not yet activated a tool like Wireshark. As the issue only appears on the production server, I'm not quite sure I will be able to find back the correct log entries, as I guess it will be massive, because of all the production activity going on, no?
So what were the results of your tests on the app servers ? Does it take 3 minutes or 1 second ?
For Wireshark, I was thinkning you could install it on your workstation, capture the traffic during the download of an attachment from your prod server then from your dev server and compare them.
Still on your workstation, does the WUT behave like your browser/MT ?
well, saving the attachment from the user tool on the test application server to a directory on that application server, took 1 second.
doing the same on the production server took 3 minutes.
the WUT behaves identical to the MT
I'll do the Wireshark test and come back to you
That shows the problem is somewhere between your AR System and DB server doesn't it ?
To check if it's at DB level or the network configuration, you could do a file transfer from your DB server to your AR System server, with scp or cifs depending of the OS.
Werner, would you have special storage parameters defined in your ardb.conf for the LOB fields ? Maybe a dedicated tablespace with specific datafiles on a distinct LUN ? The problem could be with this LUN ?
The test above (transfer via scp/cifs) will give us a hint.
Nope, no ardb.conf configured.
The transfer test could be useful still.
Also, when you see the file downloading with your browser, does it show a good but decreasing transfer speed, or just a plain slow transfer ?
After clicking ‘Save to Disk’, when looking at the directory where the fill will be downloaded to, you can see it increase it size by 100kb each 10 à 15 seconds.
What about the simple transfer test ?
Do you have a lot of simultaneous users of this platform ? I'm asking because I wonder how an AR System server can run correctly with a 10kb/s transfer rate with its DB :/
I do not have access to the oracle server, so I’ve asked the dba to do the transfer.
I’ll have to wait for his return.
The network should be a gigatebit line.
Currently, 43 users are connected.
Can you use a simple "network meter" to see the bandwith used at any given time? As Sylvain said, perhaps it's just downloading at 10kbps...
You could even use the "task manager" from Windows on the desktop, "performance" tab, you have a "ressource monitor" button (well, depends on version) and you can see a bandwith rough usage.
I know it's been a while, but I promised you feedback, so here I am.
A lot of investigation has been done during the last months.
Yesterday we finally discovered the reason.
The bad performance was just due to the fact that the an Intel network adapter was used instead of the VMware PCI Ethernet Adapter.
I think it's safe to assume that the vmware adapter is optimized to be used with vmware virtual machines, and that the Intel adapter is not
I do not exagarate, I think, when I say that overall user speed increased by x 10.
That's good news !! Congrats