[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [K12OSN] Thoughts for my upgrade...



Steve Wright wrote:

I have been working on the idea of 'simplifying larger installations' by abstracting the Servers and/or Services. I am nowhere near a complete concept, let alone a working solution, which is a pity, considering you are requesting exactly what I am building. Also, unfortunately, I have to postpone any serious Linux work for the moment while I am busy with my peak-season of employment...non IT related. 8-/

As K12LTSP gets moved out of the lab and across the campus, this becomes more and more of an issue. Right now K12LTSP is an amazing solution for a lab where it can look to the rest of the network like a single node, but I'm going to be doing serious tweaking to get it working in my environment. This is certianly not a dig on the project - it simply points out where there's work still to do.


RAID1 is mirroring ? Do you really need 4 drives as a mirror ? The Array will only be the size of a single drive, not the sum of the drives..

You're right of course - I meant RAID5, striped with parity. The capacity then is N-1, where N is the number of drives in the set. In my case, I have 4 35GB drives which gives approximately 105GB of storage. (I say approximately because /boot for example cannot be on a software RAID5 partition, so it'll be set up with RAID1 on two of the drives for mirroring, and I'll use the equivalent space on the other two drives for swap.)


User and group info will be synched up using rsync, which is how I currently manage them. I then only have to add users and groups to SVHSAPP01. It works great, and is simpler and more robust than using NIS, though admittedly not realtime. However, I'm synchronizing now every two minutes, and it seems to work fine.

Do you have a reason why not to use LDAP to achieve this ?

Lack of time to learn something new :) I would actually like to learn more about LDAP, and if anyone wants to recommend a book or website on how to set up a robust LDAP directory for several apps to use (authentication from both Linux and Windows systems, integration with our mail server, which can authenticate against an external LDAP server, other internal web based apps built in php which could use LDAP for auth, etc.)


I'm wondering if there's any reason I shouldn't export /usr, /opt, and /tftpboot from SVHSAPP01 to the other servers as well. In theory, none of those filesystems should be having anything written to them under normal circumstances. It would certainly make upgrading simple, as I'd only have to touch one machine.

I would suggest that the Terminal Servers' performance will suffer by reading its' binaries over NFS. IMO, each Terminal Server will be better to have its' own local binaries and libs.

I hear ya. If I can stick with purely rpm apps, I could probably use scp to push out new rpms and then ssh to run the rpm -i command. This would be pretty easy to script, now that I think about it. It falls apart a bit if I need to build apps from source. In reality, I'd still much rather upgrade 6 app servers by hand than 140 Windows workstations.


Yes, I have been unable to find a proper load balancer for XDMCP. That is, one the actually selects and allocates the least-loaded Server.

I cannot picture how to 'fail over' a client session. Currently, the User will lose their session, and be forced to logon anew. It may even result in a support call since the User will experience a frozen console. This is only my perception, and perhaps the clients' X Server will re-paint the 'login' screen almost immediately, or perhaps after a few mouse-clicks.

If the app server a client is connected to dies, that session is going to die. I don't see anyway around this. The failover I was describing would happen when say SVHSAPP03 dies. Every 6th boot, the DNS server would give out the IP for SVHSAPP03 to the terminal, which of course would fail. A simple reboot of the terminal would give it another chance (at 5:1 odds) of getting a good server. Ugly, but it would work.


I had been considering mosix, but the majority of our users will be using Mozilla and OpenOffice most of the time, and since those don't migrate, we wouldn't see much balancing.

Have you considered using seperate Servers for these Apps ? Mozilla and O-O run well on their own, seperate machine. So you have a Terminal Server, Mozilla Server, and an OpenOffice Server..

'Tis a thought.


All the best for your upgrade. Do keep us informed.

Thanks for taking the time to reply. I'll let you know how it goes.


--
Chris Hobbs       Silver Valley Unified School District
Head geek:              Technology Services Coordinator
webmaster:   http://www.silvervalley.k12.ca.us/~chobbs/
postmaster:               chobbs silvervalley k12 ca us





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]