Very flaky client operation

[Doug Bissett]

I have SAMBA installed in 4 different systems, and two versions of eCS in each (eCS 2.0, and eCS 2.1). I use SAMBA server (I have tried various versions, with the same results), but the server does not seem to be the cause of the problems. If I use indows (7 or XP) as the client, I don't see these problems, so I assume that it is the eCS client that is causing trouble.

A typical scenario is: Open the client, and wait for the Network neighbourhood to populate (usually happens in about 20 seconds, or never). Refresh works about 1% of the time, even when the window is already populated.

I normally use a saved config to attach things. That works as long as everything is set up properly (the usual case).

Okay, now I have a drive object (Q:) to work with. When I open that, it shows the drive, and I need to expand the window to see more. If I try to do that too quickly, nothing happens for a few seconds (or sometimes never). When it does work, I can now see the share name, with a Plus sign that can be expanded (again, as long as I wait for a few seconds before trying it). My shares on each machine are for each drive. When it works, the drives appear, usually withing about 5 seconds. Sometimes, it shows part of them, and the system is locked. During the time that it takes to display the shares, I am locked out of the system. When it fails to complete the display (about 50% of the time), the only way to recover, is to force a reboot.

Okay, if I get far enough to show all of the drive shares, I usually want to open the largest drive (about 250GB, with about 155 GB free). I click the Plus sign, and wait. Very often, it shows the first directory, with a connecting line to the second position, but no second directory. It sits there for about a minute, and continues, but only sometimes. While it is in that state, there is no IP activity, very low CPU usage, and everything is locked. About 10% of the time, it never continues, and I must force a reboot. If it does continue, it will pause a couple of times before completing the directory listing. Sometimes stopping forever when it does pause. That means that I get a complete directory listing, with a still usable system, about 70% of the time. The rest of the time, a forced reboot is the only way out.

This is most annoying, and if I am in a hurry, and don't wait for it, it almost always hangs the system. This also happens when trying to connect to the local SAMBA server (on the same machine), so it doesn't seem to be a problem with the network, and other things (FTP server, RSync, etc.) work as expected.

From the SAMBA client Help-> About: Samba Client Versions: File system EVFS Version 1.002 Plugin file M:\ecs\dll\ndpsmb.dll Plugin Vendor Netlabs.org Plugin Revision 2.01.1.2 Plugin Date/Time 15 Mar 2012 16:23:43 Plugin Build Machine HerwigB Plugin Language Code en Plugin File Version 2.1 Plugin Description NDPSMB - SMBFS NetDrive External Plugin Build GA Commandline utility path M:\ecs\system\samba Commandline utilities Version 3.3.16-eCS 1.1.1-675 EVFSGUIVersion 2.1.3 [2011-09-13]

This seems to be the most reliable combination.

What can I do to assist you to get this fixed? Thanks...

[Doug Bissett]

After being pointed to #178, I added the line:

name resolve order = bcast host lmhosts wins

to the smb.conf file. This seems to make a huge difference. The time to access the directory structure is a LOT quicker. I need to use it more, to see if it really does fix the problems, but it looks promising.

[Doug Bissett]

I followed a suggestion to use evfsgui-25beta-20120414.zip. I attempted to mount a new share, and got an error. I will upload EVFSGUI.err.

[Doug Bissett]

This is an update to the EVFSCUI error. Reference to the thread at os2world.com ​http://www.os2world.com/component/option,com_smf/Itemid,63/topic,4785.msg25138/#msg25138.

I updated my Lenovo ThinkPad T510 to use EVFSGUI-25beta-20120414. I then attempted to mount my brother's Win98 machine (DUMBER) share (C). I got the popup shown in EVFSGUI_ERR.jpg, and EVFSGUI.err. There is no .evp file, so I just clicked Mount, and got the error. There is also no smb.conf file in Win98.

This eror happens on every attempt to mount a share, using the Mount button, no matter what server, or OS, I attempt to mount.

[Doug Bissett]

I should comment on attaching to my brother's antique machine. It is an old IBM 450 mhz pentium, with just under 400 MB of main memory. If I use EVFSGUI 2.13, to attach to the Win98 system, using the "name resolve order = bcast lmhosts wins host" line in smb.conf, it works very fast, and with no hangs. I have installed eCS 2.1 on that machine (yesterday), and updated that to EVFSGUI 2.13, and the same SAMBA server as the other machines. If I then attempt to connect to all of the shares on the SAMBA server, it always hangs. If I select the largest share (drive, which has roughly 60,000 files on it), and use the default EVFSGUI settings, I get hangs. If I make the EVFSGUI settings maximum values (600 seconds, and 80 for the cache), I can actually use the share, but I still see long pauses when I first attempt to populate the tree. If I get anxious, and attempt to do anything while the tree is populating, the machine will freeze, requiring a reboot (it will do that sometimes anyway). It appears that those two settings need to be 1200 seconds. and 1000 for the cache, to make it work reliably. Faster machines seem to be able to handle smaller numbers, but not all that well.

I am very time, and resource, limited, at the moment. I should be home in one week, and then i will have more time, and my own machines to work with.

[Doug Bissett]

I tried evfsgui-25beta-20120529-update.zip, and it still crashes when I click the mount button. New evfsgui.err file uploaded.

[Doug Bissett]

I am now using the contents of evfsgui-25beta-20120603-update.zip. Other than a false error popup when I try to mount a share in the Network neighbourhood window, this seems to work okay, when I use the maximum values for "sec. cache timeout" (600) and "listings to cache" (80). Trying to use the default values (10 and 32) almost always results in the client hanging the whole desktop, while trying to populate the directory list (reboot is required).

I find it very difficult to describe some of the things that I am seeing. For instance, if I select various machines using the Selected-> Information menu item, I find two (of 4) machines claiming Role = DMB (I assume that means Domain Master Browser). The other two have Role = workstation. Then, I turned off one of the DMB machines, and nothing worked for about 10 minutes (possibly the sec. cache timeout setting?). Then, Network neighbourhood repopulated, including the machine that is now turned off. Selected-> information shows the domain, and status OK, with the rest being blank. Two of the three machines that are still on show the same thing. Only the local machine shows the expected information (Role = workstation). I can mount a share on another machine (using a *.evp file), but the tree is very slow to populate. I also can't seem to get the menu item File-> Reset LMHOSTS to do what I would expect it to do. After some time, it does create the file, listing only the local machine. The observation is, that one cannot use LMHOSTS, unless there is one machine that is always on, and is designated as Role = DMB. It also seems that Refresh, in Network neighbourhood, doesn't work properly (the machine that got turned off is still listed).

All of this is very slow, even with a gigabit wired connection. I will add more when I find some time to work with it. meanwhile, I would like some serious guidance about what settings that I should be using. Note that I do add windows (98, 2K, XP, and 7), as well as warp4, to the mix sometimes (both real, and virtual, machines), so that also needs to work.

[Doug Bissett]

Okay, now I am using evfsgui-25beta-20120605-debug.zip.

I updated three machines to this level, turned on Debug, and Logging, in the Global settings tab, and closed the program (all three machines).

Then I turned all three machines off (powered down).

Then I turned on my main machine, known as IREBBS7. Initially, NMDB.EXE started, and terminated (this will be the subject of another incident). I did a restart on the server, and it started okay. I then started the other two machines, and verified that the server started properly - they did.

Then, I started EVFSGUI on IREBBS7. Nothing showed up in Network neighbourhood (IREBBS7, ROAMIN2, and ROAMIN3, should have showed up). See evfsgui.log1.

After looking this over, I tried starting EVFSGUI on IREBBS7 again. This time Network neighbourhood populated properly, and quickly. see evfsgui.log2

Now, I selected ROAMIN3 in Network neighbourhood, and clicked the Mount button. See evfsguierr.jpg for the result.

I can find no trace of an additional log file (expected because I turned on logging), other than an empty file in %temp%. Where would I look?

[Herwig Bauernfeind]

The additional (plugin) logs are %LOGFILES%\log.ndpsmb and %LOGFILES%\log.smbc

[Doug Bissett]

I have been experimenting with various options. If I can convince one system to remain as the DMB machine, and don't change that, I get pretty good service from the SAMBA client on all machines (although WinXP and Win7, don't seem to be as responsive as they are if they can take the DMB status). If I ever reboot the DMB machine, it takes a couple of hours before things start to work properly again.

[Doug Bissett]

Okay, here is a procedure that always causes problems, for me:

Attach a USB disk drive to a machine (call it the server), and add that to the share list. Go to another machine (client), and attempt to connect to all shares on the server. It will work, most of the time. Now, disconnect the client from the server, and remove the USB device from the server (leave the share as it is). Then, try to connect to all of the shares on the server. The client will hang on the missing share. The hang blocks the desktop on the client, and a forced reboot seems to be the only way out.

You can also block the client if you restart the server while the client is attempting to open a share in the pseudo drive. Again, the only way out is to force a reboot.

[Doug Bissett]

I believe that I have found the main problem.

If I use the system with a Realtek 8169 NIC (happens to be my main system, so it is almost always one of the machines involved), SAMBA GUI is very flaky, and it doesn't seem to matter if it is being used as the server, or the client machine. I am using the Multimac driver from r8169-0.2.1.wpi, but this has been happening since I was using the GENMAC driver. When I use a virtual machine (VirtualBox-4.2.6-82870-Win.exe), using the Multimac driver from e1000e-0.2.3.wpi, through a Realtek NIC (The host is windows 7 pro, with an Intel gigabit NIC emulated for eCS), to my old Asus A8N-SLI machine, using Multimac driver nveth-0.1.10.wpi, I get reasonably good results, including being much faster (although there seems to be a problem with VirtualBox sometimes).

Other things seem to work okay. I use RSync with this machine, with no trouble, and other network programs seem to work okay (FTP, Firefox, PMMail).

[Silvan Scherrer]

if it's really a network driver problem, we can't do anything about it. and if i interpret your above comment. it leads to that conclusion.
do you have another box to try client or server?

[Doug Bissett]

I don't know if it is a driver problem, although it does appear that the Multimac driver may have a problem. I don't seem to have problems with other network programs. RSync 3.09SHL seems to work okay, with large numbers of files. I also don't have a problem when using Win7, or WinXP as the client to the machine with the Realtek NIC. I also don't have a problem with using SAMBA between my IBM ThinkPad T43, using GENMAC to drive the Broadcom gigabit NIC to my Asus A8N-SLI with NVIDIA chipset, and the Multimac NVETH driver. All of my machines have trouble using the SAMBA client to the machine with the Realtek NIC. The machine with the Realtek NIC also has trouble running the client to any of the other machines. Since it is only the SAMBA client that seems to have trouble, I suspect that there is something in the client that is making the problem show up.

Hope this makes sense...

[Doug Bissett]

Why did you close this? It is a problem in the SAMBA client, that needs to be fixed.

It seems to me that there are timeouts that are way too short, and the NIC runs slower than it should, so the timeouts get triggered.

FWIW, the client that is in eCS 2.2 beta seems to be a little better, but it still has problems.

It doesn't matter if the machine with the Realrek NIC is used as the server, or as the client, as long as a Realtek NIC is involved, the SAMBA link runs a LOT slower than it should (even if there are errors, which doesn't seem to be the case). If I use Windows (XP, or 7) as the client, the performance is very much better (by a factor of about 10). There is NO WAY that it should take 5 minutes just to populate the directory list in the network drive, when windows takes 20 seconds. The really BAD part is that if I try to do anything else while the directory list is being populated, the client machine locks up, and it is necessary to use the reset button to recover.

Totally unacceptable.

[Herwig Bauernfeind]

Doug, basically I agree with you, BUT please provide the additional plugin logs %LOGFILES%\log.ndpsmb and %LOGFILES%\log.smbc

Without these we have ZERO chance to find a bug, or find out whether Samba is involved or not, and that is why I closed the ticket with "feedback pending"...

In case you provide these, the Ticket will be reopened.

[Doug Bissett]

At the moment, this is working much better. I have updated to the recently available evfsgui-25RC4-20130409.zip. I will monitor the problem.

[Herwig Bauernfeind]

Please note that the version of EVFSGUI is completely unrelated to your problem: The plugin is a possible culprit and it is not clear which version of the plugin you are actually using.

[Doug Bissett]

From Help-> About: Samba Client Versions:
File system EVFS Version 1.002
Plugin file C:\ecs\dll\NDPSMB.dll
Plugin file C:\ecs\dll\InstallPackages\ndpsmb\ndpsmb.dll
Commandline utilities Version 3.0.37-eCS 1.0.6-565
EVFSGUIVersion 2.5.RC4-[2013-04-09]

The plugin is from ndpsmb-2_1_3-GA-20130407.wpi

I was using an earlier version until a few days ago: ndpsmb-2_2_0-beta2-20111125.wpi