Intermittent slow access to cifs mounted shares – Help stop my school going Windows

[AdamK1]

Hi,

I am having major issues with slow access to windows shares that are mounted using cifs.
The shares get mounted correctly and you can navigate through the directories and open files. The only problem is that it randomly starts going really slow taking 30 seconds or longer to open a directory that has 2 or 3 files in it.

I have tried quite a few things to try and fix this without any luck. Its getting to the point where I am having to consider recommending that we use windows instead, which I would rather not do as I think its good for students to experience different operating systems during school.

Below I'm going to describe our setup and things I have tried, so if you have any idea as to why we are experiencing this problem, or having an ideas on how to fix it or even ideas on how I can improve the setup, then please reply to the thread.


Setup:
Windows 2008 server with Active Directory authentication. This server also holds the users home directory shares and all other resource shares.
Fedora 10 NFS server for Linux profiles – Authenticates against Windows 2008
4 classrooms of 30 computers – Fedora 10 running GNOME – Authenticates against Windows 2008
Roughly 1000 students, each with a Linux profile and Windows home share.
Gigabit fibre connection between server room and a switch cab that connects 3 of the classrooms, 100Mb fibre connection between server room and cab for the 4th classroom.
Cab for the 3 classrooms has 5 switches, 4 switches uplink to the one main switch that then has the fibre connection – all 5 switches are Gigabit HP Procurve.

All computers mount the NFS share for the Linux profiles so that we can lock down their profiles and easily manage any changes. The share is mounted in /home/DOMAIN/ and is done on boot using fstab

When a student logs in, pam mount mounts the following shares:
Home Share: /home/DOMAIN/username/HOME
Student Resources: /home/DOMAIN/username/Student Resources

Each user has their own .pam_mount.conf.xml file in their profile so we can allow access to other shares where needed.

Things I have tried:

• I monitored the processors on the NFS server as it isn't a high spec server and noticed it spiked now and then, so we setup a better server and experienced the same problems – the processors are not spiking with this server though.
• Stopped using GNOME Nautilus as the file manager and started using Dolphin, I think this might of reduced the frequency of the problem, but it still happens regularly.
• I thought the problem might be that the shares are mounted in /home/DOMAIN which is also a mounted share. So I modified everyone's pam mount file to mount the shares to /mnt/username/sharename but the problem has still continued.

I've thought there might be too much load on the main switch for the 3 classrooms, but that wouldn't explain why the 4th classroom still has the same problem and the test machine that I have in my office has had the problem too.

I think I can rule out the NFS server as the cause as the response time for accessing that is fine.

The windows 2008 server is way overpowered for what it does, really we should have it virtualised and running 4 servers on it, so I think I can rule that out.

I'm wondering if its the cifs protocol, is there a different one we could use to mount the windows shares?

I have also wondered if its just Fedora, would anyone suggest using a different distro instead? Ideally I would like to solve this problem using fedora as its a big task having to setup a different distro then image all the computers.

If anyone needs any more information to help me, then just ask.

Thanks,
Adam

[rituraj.goswami]

thatz strange, the cifs protocol shouldn't be a problem. i've implemented the same thing but it's working fine....!!!!!!!!

[AdamK1]

Well since my original test I have ruled out GNOME being the problem as we got the same results using KDE. I'm starting to think it might be Fedora 10. I have a similar setup with my machine in the office using Fedora 11 and I don't have the problem but I found out last week that my colleague who uses fedora 10 does have the problem.

Currently setting up a test machine in Fedora 11 and will set one of the teachers up with it.

So fingers crossed!

Where you say you have the implemented the same thing Rituraj, did you do it in the same way I did, or slightly different, would be interesting to know.

Adam

[Rubberman]

When this happens, are all the clients slow? If the clients (or servers) are using DHCP, what is the time-to-live for the addresses they are assigned?

[AdamK1]

Sometimes over half a class (15+ students) will notice it, at times, its just the odd one here and there. But it can be at several different points during the day. The servers have static addresses, whereas, the client are on a 24 hour lease.

I have a Fedora 11 machine set up now for testing. Will be putting it into a classroom today for a member off staff to use. So hopefully she doesn't notice any slowness.

Adam

[Rubberman]

Ok. I re-read your original post and think I understand now. You are using a Windows 2008 Server system as a file server, and your client/student systems are running Linux (Fedora?), connecting to their "home" directories using a CIFS mount. A couple of things you don't mention are:

1. What is the hardware configuration of the Win2008 server (CPU's, cores, RAM, disc types and configuration (sata, ide, raid0-N, etc), number/type of ethernet ports).
2. What is the router and what is the pipe/bandwidth between the router and switches? Or is one of the HP switches doing double-duty as a router?

[AdamK1]

Ok. I re-read your original post and think I understand now. You are using a Windows 2008 Server system as a file server, and your client/student systems are running Linux (Fedora?), connecting to their "home" directories using a CIFS mount. A couple of things you don't mention are:

1. What is the hardware configuration of the Win2008 server (CPU's, cores, RAM, disc types and configuration (sata, ide, raid0-N, etc), number/type of ethernet ports).
2. What is the router and what is the pipe/bandwidth between the router and switches? Or is one of the HP switches doing double-duty as a router?

I'm at home at the moment so can't give full spec, but its way over powered for what it does.

Off the top of my head:

HP
G5
Dual Quad Core Processors
8GB RAM
About 750GB storage
RAID 6 setup
4 Network cards (1 Gb) - Only using one of the ports though.


The server is our domain controller, so it does all the user authentication and stores all their home folders and any other share.

Our school has Linux, Windows and Mac, so we decided having a Windows server for authentication would be the easiest solution so that we can tie all OS's together.

All the servers and clients are connected together via HP Procurve switches.
Most of the clients passthrough a Gb switch then over Gb fibre, back to another Gb switch then to the server. However, I think the network can be ruled out as I have seen the problem with a machine that is connected directly to the same switch as the server.

The thing that confusses me the most is that it doesn't happen all the time, so I can be testing a change I have made and it seems to work fine, but once I have 90 - 120 students using the machines people randomly start having issues.

Also, I have never experienced the problem with my Fedora 11 machine and I'm on that all day, but then again, I tend to use terminal a lot.

Adam

[rcgreen]

What do you use for name resolution? This is a common cause of slowness in mixed networks.

[AdamK1]

The same Windows server is also the DNS server and is set as the primary DNS on all the clients.

I get fast responce times when pinging host names from the clients too.

[AdamK1]

The fedora 11 machine I set up doesn't seem to have solved the problem either.

I have just changed the default kernel on the Fedora 10 machines to see if that makes any difference. Can't see it doing so, but I'm all out of ideas now.