Problems with NLB

mr2nut

I've tried to NLB using two Virtual Server 2003 SBS Servers. I know you shouldn't really do it on domain controllers but it's only for test purposes so should work ok right? Do I have to set multicast on each Server under the NLB tab or does it need to stay on unicast? It says something about not being able to connect to all nodes if unicast is on when you load NLB manager up. It keeps going to converged then pending and now it's on miconfigured lol. Any ideas as to why it's not working?

jbaello

This might be a good resource for you, it's been awhile since I configured NLB when studying 293, but I used this link, and there are steps here to configure one.

http://technet.microsoft.com/en-us/library/cc775749.aspx

royal

Good timing. I made a post out to the Exchange newsgroups yesterday in regards to Unicast Vs Multicast Vs Multicast IGMP.

Unicast overwrites the MAC. So because of that, you need 2 NICs when using Unicast. The NICs dedicated to Unicast will be used for incoming traffic. These NICs should be on their own Vlan, Switch, or a hub off of a switch so it doesn't turn your switch into a hub. This is because the two NICs will have the same MAC and because of that, the switch will end up sending data to every port on the switch turning it into a hub.

For multicast, it doesn't re-write the MACs, it essentially adds the MAC to the existing NICs. Because of this, you can use 1 NIC. But your switch needs to be able to configure static ARP entries so the switch knows which server is which. If your switch supports IGMP, you can enable Multicast IGMP which your switch as well as your NICs need to support. I believe IGMP will automatically prevent switch port flooding.

Personally, I would just use two NICs and Unicast Mode.

mr2nut

royal wrote:

Good timing. I made a post out to the Exchange newsgroups yesterday in regards to Unicast Vs Multicast Vs Multicast IGMP.

Unicast overwrites the MAC. So because of that, you need 2 NICs when using Unicast. The NICs dedicated to Unicast will be used for incoming traffic. These NICs should be on their own Vlan, Switch, or a hub off of a switch so it doesn't turn your switch into a hub. This is because the two NICs will have the same MAC and because of that, the switch will end up sending data to every port on the switch turning it into a hub.

For multicast, it doesn't re-write the MACs, it essentially adds the MAC to the existing NICs. Because of this, you can use 1 NIC. But your switch needs to be able to configure static ARP entries so the switch knows which server is which. If your switch supports IGMP, you can enable Multicast IGMP which your switch as well as your NICs need to support. I believe IGMP will automatically prevent switch port flooding.

Personally, I would just use two NICs and Unicast Mode.

Thanks for getting straight to the point "I would just use two NICs and Unicast Mode" That's the kind of information you can't find when searching through google and having to read the pages and pages of explanation and theory lol. I'll create another virtual NIC on my Virtual PC Server for both Servers and test again tomorrow. Should work a treat with both Domain Controllers using Unicast with 2 NICs each shouldn't it?

Does there have to be any specific ranges of IP Addresses on each Servers NIC like Clustering where each IP range and subnet has to be different?

Also, with NLB, am I right in thinking that as it is not as safe as Clustering (i.e. one Server goes down, the other nodes can reach the central storage to continue access) with NLB, if the main Server goes down that you set up the NLB on, does that screw everything up, or does NLB put some sort of DNS record on the Server that somehow re-directs clients to the alternative 'Web Servers' if you like?

(Another quick q ) How exactly does NLB distribute traffic? Does it work similarly to round robin in DNS where each query that comes in gets evenly distributed to a different Server for each request?

mr2nut

Does this look right to you Royal?

Diagrams really help me to understand and visualise things as I haven't covered this stuff before.

mr2nut

Network load balancing works the same way a cluster works in that, you have for example 2 nodes connecting to an external storage device that hosts your website? So what I would need to do is install IIS on both NLB Servers and link IIS to the external storage, so that if one Server went down, people could still access the website? Sometimes I think I have grasped this, then when I come to trying it on my virtual lab I don't know what to do lol.

bertieb

mr2nut wrote:

Network load balancing works the same way a cluster works in that, you have for example 2 nodes connecting to an external storage device that hosts your website? So what I would need to do is install IIS on both NLB Servers and link IIS to the external storage, so that if one Server went down, people could still access the website? Sometimes I think I have grasped this, then when I come to trying it on my virtual lab I don't know what to do lol.

Typically in NLB farms you don't have a shared storage device. For example, in a typical IIS scenario you'd have IIS running on each server, configured the same, and you would load-balance the required ports but you would have identical copies of the data/web application files on each node. You can use any number of methods for keeping each server in sync whether its a manual process or using DFS, robocopy, xcopy etc. Of course, you can configure IIS to use shares to link to a central location but I've not seen it that often. Think of NLB as a means of distributing connections (load) between multiple devices rather than it using shared storage - it doesn't require it whereas clustering does.

Data that changes frequently is often stored in a database, and of course you'd use clustering as a high availability technique for that

Hope I've explained this OK, I've just re-read it and it's as clear as mud to me though I'm v.tired

mr2nut

That explains it perfectly thanks. All this clustering and nlb stuff all at once, it can really get confusing at first until you can grasp it all. The problem I find is that there's so many resources on the net that go into far too much detail and it's hard to pinpoint the important stuff. I did try to look for the easy answers before asking here so i'm not been lazy :P

mr2nut

I now have IIS on both machines and have been looking into mirroring the data from the main Server every say, 15mins for a small website. I've tried robocopy but this doesn't copy open files, neither does xcopy. I have seen that ntbackup can do this. Can you simply set up an ntbackup going and set it to copy data every 15mins? Would this be a good solution or is there a more preferred way?

sprkymrk

mr2nut wrote:

I now have IIS on both machines and have been looking into mirroring the data from the main Server every say, 15mins for a small website. I've tried robocopy but this doesn't copy open files, neither does xcopy. I have seen that ntbackup can do this. Can you simply set up an ntbackup going and set it to copy data every 15mins? Would this be a good solution or is there a more preferred way?

If you can take the website offline for a minute, let robocopy do its thing once, then set it to run every 15 minutes will that work for you? How often is the website updated? Otherwise yes, ntbackup should work, but unless the site is updated almost constantly why would it need to run every 15 minutes vs once per night or something?

You could use schtasks to run an ntbackup batch file every 15 minutes something like this:

schtasks /create /sc minute /mo 15 /tn "WebsiteBackup" /tr C:\Tools\ntbackup_website.bat

However, even at that ntbackup wants you to create a "bkf" file, not just a direct copy like you would get with robocopy or something.

dynamik

With robocopy, is your problem with copying files that are open or copying over files that are in use? If it's the former, just use a source, such as your local disk, to copy to both destinations. If it's the latter, you can adjust the wait and retry settings, and the destination file will probably be available eventually, especially if it occurs during a period of low activity.

Instead of having to screw around with scheduling, why don't you just create a script and run as needed?

undomiel

You could also use VSS to work with robocopy to keep things in synch.

http://www.eggheadcafe.com/tutorials/aspnet/f6972828-1e81-4cd4-ae0c-36196a82ed25/workstation-open-file-bac.aspx

mr2nut

sprkymrk wrote:

mr2nut wrote:

I now have IIS on both machines and have been looking into mirroring the data from the main Server every say, 15mins for a small website. I've tried robocopy but this doesn't copy open files, neither does xcopy. I have seen that ntbackup can do this. Can you simply set up an ntbackup going and set it to copy data every 15mins? Would this be a good solution or is there a more preferred way?

If you can take the website offline for a minute, let robocopy do its thing once, then set it to run every 15 minutes will that work for you? How often is the website updated? Otherwise yes, ntbackup should work, but unless the site is updated almost constantly why would it need to run every 15 minutes vs once per night or something?

You could use schtasks to run an ntbackup batch file every 15 minutes something like this:

schtasks /create /sc minute /mo 15 /tn "WebsiteBackup" /tr C:\Tools\ntbackup_website.bat

However, even at that ntbackup wants you to create a "bkf" file, not just a direct copy like you would get with robocopy or something.

I suppose syncing the website isn't required every 15mins, I was just using that as an example as I didn't think about open files been a problem to be fair. The site will remain static mostly so just once a night would be fine. It sounds like robocopy would suit my requirements fine then.

If Robocopy sees that files are open, will it simply skip those files until it's next backup when it will retry and then copy them if they are inactive? Can Robocopy run as schedules? If so, do you need to set up a batch file and a scheduled task, or is there a more preferred method?

mr2nut

dynamik wrote:

With robocopy, is your problem with copying files that are open or copying over files that are in use? If it's the former, just use a source, such as your local disk, to copy to both destinations. If it's the latter, you can adjust the wait and retry settings, and the destination file will probably be available eventually, especially if it occurs during a period of low activity.

Instead of having to screw around with scheduling, why don't you just create a script and run as needed?

I'm not sure I understand what you mean here dynamik. Isn't a file open the same as a file in use?

mr2nut

undomiel wrote:

You could also use VSS to work with robocopy to keep things in synch.

http://www.eggheadcafe.com/tutorials/aspnet/f6972828-1e81-4cd4-ae0c-36196a82ed25/workstation-open-file-bac.aspx

Although that looks very clever, after such a hard day as the office that was far too much scripting to get my head around Think i'll go for just a daily sync with Robocopy, it sounds like the simplest way.

dynamik

Yea, but I was asking if it was the source (reading) or destination (writing) that was giving your problems.

mr2nut

dynamik wrote:

Yea, but I was asking if it was the source (reading) or destination (writing) that was giving your problems.

I see, sorry, I didn't read it properly. I've been so busy today that my posts I have checked on I have more glanced over.

mr2nut

Well, i've done it and it's working fine. Cheers for the tips guys. Both nodes are showing up as converged and i've set up a batch file with scheduled tasks to run on a dialy basis with an /mir switch. Works a treat!!

mr2nut

Might as well bring this NLB thread back up to post for another NLB query..

Now that I have mastered the multicast way of setting up NLB, I was looking into going down the unicast route as I have heard that it is a good idea to split NLB traffic from your public facing traffic.


I have 2 servers here with 2 NICs each set up as follows..


Server1.testdomain.local Server2.testdomain.local

Private = 10.0.0.1, 255.0.0.0 Private = 10.0.0.2, 255.0.0.0
Public = 192.168.1.1, 255.255.255.0 Public = 192.168.1.2, 255.255.255.0


I have set it up this way as i've read that having your public traffic on a seperate subnet is good practise?

Now i'm trying to look for a very thorough guide as to how to set up unicast NLB, but all I can find are vague answers and no detailed guides with pictures and clear explanations of not just where stuff goes, but also why. If anyone has a link kicking around or has some spare time to do this, it would be great.