what could my boss be talking about?

Smallguy

some of you probalby recall my thread about Clustring and a disagreement with my boss.

anyways I sent him an email explaining what clustering is and why the plan he had to use the RAID 5 on the new exchange server and have the old one as a paasive node will not work (with out seperate attached storage)

so ot day he asked me ot look into it again so I asked ihm if he wantd me ot price up a SAN and go form there he told me that a Cluster doe;snt need seperate storage ot to dig deeper because there is a way to do a cluster and not have sperate storage but to have exact replica's of the data opn each server. One server being active and the other being passive. that way you are ont relying on even a single San Failing but the boxes are completly iondependant of one and other.

anyways this is not something I've ever heard of so either my inexperience is catchnig up or terminology is being confused.

so does anyone know of a way to create redudancy where each server has the exact same copy of the infomation so if one fails the other just will take over while keeping the data synced between the 2 machines so there is no single point of failure.

davenport

I think I know what you mean. Is he talking about something along the lines of round robin DNS with two servers? That would provide load balancing but no fault tolerance.

royal

davenport wrote:

I think I know what you mean. Is he talking about something along the lines of round robin DNS with two servers? That would provide load balancing but no fault tolerance.

This has nothing to do with distributing/load balancing client traffic. Your cluster has its own DNS Host Record in which clients point to. The cluster service will then have the active node take care of the request. This has to do with replicating traffic across Direct Attached Storage (DAS) and having each cluster be able to use its own storage when it is active. Clusters typically require external shared storage to work.

Personally, I have not heard of being able to do such a thing. I don't have much experience when it comes to clustering, however. I do know that with Server 2003 Enterprise Sp1 and with a MNS cluster, you can force the cluster to not require a 3rd node and have it use a file share instead. Perhaps there is some relevance. You can check out the doc at http://support.microsoft.com/kb/921181

Smallguy

THnaks for the info

unfortuantley he is talking about using 2 machines

apparently the guys the company he works for (which Im' not a part of) has researched this and what I found they agreed with intially aswell but now they have information that says you do not need a SAN or seperate storage to have 2 nodes with exchange one active one passive and if the active fials the passive takes over because it somehow has an exact replica of the data stored locally.

to me this sounds fishy and I've never heard of it in anything I've read ( but I am a newbie)

to be quite honest if they already have the info in hand I do not se why I am being asked to re-invent the wheel since they already spent hundreds of hours on researching this.

I guess it's being the new guy on the block but it can drive you crazy

sprkymrk

I found this that shows an Exchange Active/Passive node clustering example still using SANs.

http://technet2.microsoft.com/WindowsServer/en/library/9013f3cf-c801-4599-92d0-c3c331dad2911033.mspx?mfr=true

And this article seems to indicate you have to use shared storage because of the Quorum Disk Resource.

http://www.microsoft.com/technet/prodtechnol/exchange/2003/insider/clustering.mspx

Unless they are referring to a Network Load Balancing cluster rather than a Quorum Device cluster like here:

http://technet2.microsoft.com/WindowsServer/en/library/f41adff5-87df-483e-9edd-8484501d79921033.mspx?mfr=true

blargoe

He isn't talking about clustering at all. He is talking about something called replication which isn't something Microsoft currently offers but some of the SAN vendors do I think (which is probably where he heard about it). Basic idea is, you have a server sitting in standby, you set up the replication software, and it makes a complete bit for bit copy of the exchange server's hard disk to the hard disk of the standby system. Once this is complete it will syncronize every so often and change the bits on the standby system's hard disks that were changed on the production systems's hard disks.

A very cool technology, but can be very expensive too from all accounts I have gotten.

davenport

icroyal wrote:

davenport wrote:

I think I know what you mean. Is he talking about something along the lines of round robin DNS with two servers? That would provide load balancing but no fault tolerance.

This has nothing to do with distributing/load balancing client traffic. Your cluster has its own DNS Host Record in which clients point to. The cluster service will then have the active node take care of the request. This has to do with replicating traffic across Direct Attached Storage (DAS) and having each cluster be able to use its own storage when it is active. Clusters typically require external shared storage to work.

Personally, I have not heard of being able to do such a thing. I don't have much experience when it comes to clustering, however. I do know that with Server 2003 Enterprise Sp1 and with a MNS cluster, you can force the cluster to not require a 3rd node and have it use a file share instead. Perhaps there is some relevance. You can check out the doc at http://support.microsoft.com/kb/921181

I'll have to take your word for it. I can't figure out what the guy is asking for.

royal

blargoe wrote:

He is talking about something called replication

Yep. mostly.

icroyal wrote:

This has to do with replicating traffic across Direct Attached Storage (DAS) and having each cluster be able to use its own storage when it is active.

Smallguy wrote:

THnaks for the info

unfortuantley he is talking about using 2 machines

With the link I gave you, you should be able to accomplish what you need. Majority node set uses its own disks for information and for quorum data. By default, the minimum amount of nodes you need with Majority Node Set Clustering is 3. With the link I gave you, you can get a majority node set down to 2 nodes instead of 3. Now all you need to do is replicate the information between the 2 servers like I originally stated. This will allow you to have a 2-node cluster that uses its own disks, which is what you originally needed.

Smallguy

icroyal wrote:

blargoe wrote:

He is talking about something called replication

Yep. mostly.

icroyal wrote:

This has to do with replicating traffic across Direct Attached Storage (DAS) and having each cluster be able to use its own storage when it is active.

Smallguy wrote:

THnaks for the info

unfortuantley he is talking about using 2 machines

With the link I gave you, you should be able to accomplish what you need. Majority node set uses its own disks for information and for quorum data. By default, the minimum amount of nodes you need with Majority Node Set Clustering is 3. With the link I gave you, you can get a majority node set down to 2 nodes instead of 3. Now all you need to do is replicate the information between the 2 servers like I originally stated. This will allow you to have a 2-node cluster that uses its own disks, which is what you originally needed.

how is the replication setup.... links are good I'll do my own reading and ask questuiions later but the only tihng I've ever setup replication with is offline address books and public folders I've never attempted to replicate an information store and I must be googleing something wrong cause all I can find is replicating address books and public folders.

Smallguy

I've done some research and then presented it ot my boss about replication software but he says the solution is on the web and requires no third-pary software or external storage and he found it a while ago but will not give up his source or has forgotten it (not sure which)

I talked to my peers and they haven't even started research for a seperate but simular project.

I'm literally going nuts over this because I just do not see how this is possialbe with out either seprate storage or replication software.

I've googled for hours... posted here asked at msexchange.org tlaked an an IT consultant I know and I'm going in circles.

blargoe

He must be thinking about majority node set then.

TechJunky

I use replication daily for Database's for clients with more than one location using a WAN. Replication has worked great for databases purposes. The only downside, and this is due to the replication tool we are currently using... Most databases compare data from one database to the other database and then make the needed changes. However, with this replication it creates an exact duplicate of the database at one location and then the same thing happens, the two databases are then compared against a single database from the last synchronization. This causes HUGE databases file size sometimes if replication does not run atleast once a day.

We also deal with offsite backup solutions, think of it as RAID off site. Perhaps that is what your boss is referring to?

TechJunky

Here is a link of a site we use...

http://www.netmass.com/sb_serversafe_overview.html


They do Exchange CDP Backup.

Smallguy

blargoe wrote:

He must be thinking about majority node set then.

it is quite possiable I'm gonig to run it by him.

hopefully that is what he wants because with a budget of zero creating redudancy is kindda difficult

even setting up a majority node cluster with the 2 boxes will not achieve what he wants...because one of the machines will have to store the data and if that machine fails for any reason you not getting fail over to work.


I am begining ot think he is thinknig fo NLB.... but that only works well with fornt end exchage servers and backends are clustered....we do ont have frontend/backend configuration.

I think this may be one of those impossiable situation eventually unless I can get a bit of money for this project.

royal

Smallguy wrote:

even setting up a majority node cluster with the 2 boxes will not achieve what he wants...because one of the machines will have to store the data and if that machine fails for any reason you not getting fail over to work.

Smallguy, I'm curious if you even read the link I provided you or what I have been posting. I have been telling you the entire time to use a Majority Node Set Cluster with the link I gave you which shows you how to run a 2-node MNS cluster while using a file share to any workstation/server which will allow you to fail over successfully. They will both have the same data because you should have replication software replicating data between the 2 cluster node's Direct Attached Storage (DAS).

garv221

Smallguy wrote:

I've done some research and then presented it ot my boss about replication software but he says the solution is on the web and requires no third-pary software or external storage and he found it a while ago but will not give up his source or has forgotten it (not sure which)

Your boss sounds like a prick, I would fire him. What- does he think this is the Lord Of The Rings and you are on a quest of clustering? Obviously he is no hurry if he knows the answer and fails to give an example.

Smallguy

icroyal wrote:

Smallguy wrote:

even setting up a majority node cluster with the 2 boxes will not achieve what he wants...because one of the machines will have to store the data and if that machine fails for any reason you not getting fail over to work.

Smallguy, I'm curious if you even read the link I provided you or what I have been posting. I have been telling you the entire time to use a Majority Node Set Cluster with the link I gave you which shows you how to run a 2-node MNS cluster while using a file share to any workstation/server which will allow you to fail over successfully. They will both have the same data because you should have replication software replicating data between the 2 cluster node's Direct Attached Storage (DAS).

NM I'm not sure how I missed it but the file share witness asks like a third node so it allows the fail over

I kept mis-reading it....must be one of those weeks i guess

royal

Smallguy wrote:

I kept mis-reading it....must be one of those weeks i guess

We all have those days. Gotta love those times when you read the same page in a book over and over and over and you realize after the 10th time reading it, you still have no idea what that page was about.

Smallguy

icroyal wrote:

Smallguy wrote:

I kept mis-reading it....must be one of those weeks i guess

We all have those days. Gotta love those times when you read the same page in a book over and over and over and you realize after the 10th time reading it, you still have no idea what that page was about.

Yeah it was driving me bonkers...if this works i'll be a happy happy man.

I kepy reding you can't do it with 2 nodes and was missing that the update allows u to do it with 2 nodes

I eveen read a white paper *(must be outdated) saying u need 3 nodes

i'll present it ot the boss monday

and thank you for the resource....I've literally asked everywhere I can think of and you were the olny one woh found this....if u don;t mnid how did u find it?

royal

Smallguy wrote:

icroyal wrote:

Smallguy wrote:

I kept mis-reading it....must be one of those weeks i guess

We all have those days. Gotta love those times when you read the same page in a book over and over and over and you realize after the 10th time reading it, you still have no idea what that page was about.

Yeah it was driving me bonkers...if this works i'll be a happy happy man.

I kepy reding you can't do it with 2 nodes and was missing that the update allows u to do it with 2 nodes

I eveen read a white paper *(must be outdated) saying u need 3 nodes

i'll present it ot the boss monday

and thank you for the resource....I've literally asked everywhere I can think of and you were the olny one woh found this....if u don;t mnid how did u find it?

We have a distribution list in our company where over a hundred engineers/senior engineers/architects discuss issues and help each other. One day they were talking about a similar issue and one of them showed a link.

Smallguy

icroyal wrote:

Smallguy wrote:

icroyal wrote:

Smallguy wrote:

I kept mis-reading it....must be one of those weeks i guess

We all have those days. Gotta love those times when you read the same page in a book over and over and over and you realize after the 10th time reading it, you still have no idea what that page was about.

Yeah it was driving me bonkers...if this works i'll be a happy happy man.

I kepy reding you can't do it with 2 nodes and was missing that the update allows u to do it with 2 nodes

I eveen read a white paper *(must be outdated) saying u need 3 nodes

i'll present it ot the boss monday

and thank you for the resource....I've literally asked everywhere I can think of and you were the olny one woh found this....if u don;t mnid how did u find it?

We have a distribution list in our company where over a hundred engineers/senior engineers/architects discuss issues and help each other. One day they were talking about a similar issue and one of them showed a link.

sounds like a pretty valuable resource to me