I have used samba for nearly 9 years with no problems and we have about 20 users. In the past we have had a dedicated samba server. We have recently virtualized this server to a quad core Q6600 using vmware virtual server 1.0.4 on a 64 bit host running ubuntu 7.10. As part of the virtualization I installed samba 3.0.26a in ubuntu 7.10. Since this has happened users have complained about random delays. The machine itself is not using much CPU or memory. When I ssh and run top there is very little CPU utilization. I increased it from 1 to 2 cpus with no difference observed. I have noticed to things in the logs of interest: [2008/03/06 12:20:10, 0] lib/util_sock.c:write_data(562) write_data: write failure in writing to client x.x.x.x. Error Broken pipe [2008/03/06 12:20:10, 0] lib/util_sock.c:send_smb(769) Error writing 64 bytes to client. -1. (Broken pipe) Also [2008/03/06 14:29:57, 1] smbd/service.c:close_cnum(1230) yyyy (x.x.x.x) closed connection to service zzzz I noticed that when I run nbtstat on the client machine the net bios name has expired and it seems to take a while to find it again. I have samba configured as a WINS server. I disabled oplocks on this particular share but the error still occurs. I have also disabled level 2 oplocks and kernel oplocks. I have tried keep alive = 0, keep alive = 30, keep alive = 180 which seems to help a little bit with the delays. I have tried various socket options including SO_RCVBUF=8192, SO_SNDBUF=8192, IPTOS_LOWDELAY, TCP_NODELAY, SO_KEEPALIVE. At the moment I have set SO_RCVBUF and SO_SNDBUF to be equal to 1400 as I noticed the MTU of the network card was 1500 which seems to but down on the broken pipes. The server has a GB onboard network card and the clients mostly are GB network cards but some of them passthrough IP phones which are 100MB. At any time when the delay (up to 60 s) occurs I can ssh into the samba server instantly using its dns name. Aside from random delays the setup seems to work fine. -- Adam Zimmer President Arius Software Corporation (519) 885-9045 x122
On 3/6/2008, Adam Zimmer (adam@ariussoftware.com) wrote:> I have tried various socket options including SO_RCVBUF=8192, > SO_SNDBUF=8192, IPTOS_LOWDELAY, TCP_NODELAY, SO_KEEPALIVE. At the > moment I have set SO_RCVBUF and SO_SNDBUF to be equal to 1400 as I > noticed the MTU of the network card was 1500 which seems to but down > on the broken pipes.I'm not saying this is cauing your problem, but you shouldn't be setting these at all, as long as you have a modern kernel (2.6 series)... These haven't been needed for a long time. -- Best regards, Charles
At the moment I have enabled timeSync with vmware tools. In the general area of time keeping on the host, I added the following settings which avoided errors about the RTC missing interrupts: host.usefastclock=false host.cpukHz=2400000 host.useTSC=true ptsc.useTSC=true I have two other machines similarly configured (with the exception of running other linux applications not samba). Ntpdate seems to be installed as it is part of the ubuntu-server default config. However, my other machines seem to run it ok. If anything they fall behind a bit and the vmware sync keeps them up-to-date. Adam Zimmer President Arius Software Corporation (519) 885-9045 x122 Ian McDonald wrote:>>> > > How are your time sync options set for the VM? Is it keeping time ok? > (note,AFAIR, you're not supposed to run NTP within a VM.). > > -- > ian >
I checked running mii-tool and it indicated that it was only using 100basetx-FD. Not sure how to change this as using vmware server UI (just vmware server 1.04 not ESX) I don't see any options and also the config file doesn't list anything for the VM. The previous server was running 32 bit. Adam Zimmer President Arius Software Corporation (519) 885-9045 x122 Lukasz Szybalski wrote:> On Thu, Mar 6, 2008 at 3:56 PM, Adam Zimmer <adam@ariussoftware.com> wrote: > >> I have now removed those socket options. I am running Linux 2.6.22. >> >> However, the delays persist. Any other ideas? I thought it might be name >> resolution so I tried: >> name resolve order = wins host bcast >> >> But this hasn't helped either. >> > > Could you check what settings are set for the vmware network card > speed. Is it in fact 1gb? My vmware as default made a 10/100 network > card. Not sure what would be some other vmware specific settings that > could be checked to see if it has issues. Was the previous server a > 64bit as well? > > Lucas >
On Thu, Mar 6, 2008 at 10:13 PM, Adam Zimmer <adam@ariussoftware.com> wrote:> I have used samba for nearly 9 years with no problems and we have about > 20 users. In the past we have had a dedicated samba server. We have > recently virtualized this server to a quad core Q6600 using vmware > virtual server 1.0.4 on a 64 bit host running ubuntu 7.10.bad idea. Vmware server is not meant for production servers. Don't try to save a buck and buy a copy of esx. It will save you all this trouble and time is money. If you really want to go along the free road, get yourself xen, linux runs perfectly with the opensource 'free as in free beer' xensource. Vmware server is a great testing tool, not a production one. -- Groeten, J.Asenjo