File server best practises for Packaged DL380 G4 + MSA500 G2

[Iannuccelli]

Hi,
I am ready to buy a Packaged DL380 G4 + MSA500 G2 with five 300 Go disks for shared storage (each nodes will have 2 RAID 0+1 72.8 Go HD). The goal is to provide an active/active 1 TB high available file and print server to users, scalable to 4 TB.
Along this forum, I read that each node must see all storage logical drives but I also read that a disk cannot be accessed at the same time without data corruption risks ...
Can you give me a small explanation (or url) of how scsi drives are shared through MSA500 G2 hardware and finally through windows 2003 EE (using cluster service).
My current understanding is that each node must see all but only touch it own part (except during fail over) !
At the end, what are the best practises for such a file server concerning RAID5 array number and logical disks number. I also plan to use each node as domain controllers, is it possible ?

Thanks

[Doug de Werd]

There is quite a bit of information here:

http://h18004.www1.hp.com/solutions/enterprise/highavailability/whitepapers/ms-packagedcluster.html

I would look at the best Practices for the G3 paper (there is not one for the G4, but they are basically the same)

These are also pretty good:

http://www.microsoft.com/windowsserver2003/technologies/clustering/default.mspx

http://www.windowscluster.com/

http://www.nwnetworks.com/cluster.html

ftp://ftp.compaq.com/pub/products/servers/proliantstorage/sharedstorage/msag2-solution.pdf

http://www.microsoft.com/technet/prodtechnol/windowsserver2003/technologies/clustering/sercsfaq.mspx




In general, you are correct about access to the disks. Both of the serevrs must be configured (via Selective Storage Presentation) to have access to all of the LUNs in the MSA500. It is OK to create one large RAID 5 set and have multiple volumes (LUNs) within it. The cluster software will manage access, and only one server will have access to a LUN at a time.

For example, you might make a RAID 5 set and put 3 LUNs in it. One would be a 500 MB volume for the Quorum drive (Q:), and the other two would be for the file shares (call them F: and G:). Both servers would be configured via SSP to have access to all 3 LUNs. However, when you install the cluster software, you would make server 1 the "Preferred Owner" for drive F, and server 2 the Preferred Owner for drive G. During normal operations, server 1 will only see drive F and server 2 will only see drive G. If server 1 fails, server 2 will take ownership (steal) drive F, and then republish the file shares to the users.

Hope this helps.

Thanks,
Doug

[Iannuccelli]

Thanks for your quick and valuable response.
Before reading all documentation, I still see a paradox about SSP. In my head, SSP is used to mask logical units to servers, but in your explanations, you said that you use it to show LUNs to all servers.
Later in your text, you say again that all nodes see all drive letters (Q, F and G) but in normal operation server 1 will only see drive F and server 2 will only see drive G. You meant that each server only use or only see ?
Last questions, is there any limit of HD drive number in an array and what about using nodes as domain controllers ?
Thanks a lot

[Doug de Werd]

With the MSA500, there is no limit to the number of drives in an array - up to the physical limit of 14 drives that can be put into the MSA500.

It is a little bit confusing with SSP and clusters. Think of it as 2 layers. In the first layer - via the SSP and ACU - you set access to all LUNs to both servers. This is underneath the OS. This is only a requirement when doing clustering - in a non-clustered configuration you would not do it this way. The reason you must set this up is that during the process of a cluster failover, the surviving server will "steal" access to the failed servers LUNs. So SSP must be set up to potentially allow either server to own and control the LUN.

At the second level is the cluster software. Even with a cluster, the LUNs are only accessed and "owned" by one node at a time. Because of how SSP is set up, the OS and cluster software is aware that it can see the other server's LUNs, but the cluster software prevents it from actually accessing them (unless there is a failover).

So SSP makes it possible for the other server to take over the failed server's LUNs, but he cluster SW manages access so that only one server can access a LUN at any given time.

As for domain controllers, you can do it, but there are some things you should be aware of.

Look here for more information - look at the question regarding setting up nodes as domain controllers:

http://www.microsoft.com/technet/prodtechnol/windowsserver2003/technologies/clustering/sercsfaq.mspx#EGCAA

Thanks,
Doug

[Iannuccelli]

OK, I better understand the system now !
Thanks for your patience ...
To check I have fully understand (or not!), it is technically possible to use a single logical drive shared by the 2 nodes BUT it is not a good idea because of the time spent by cluster SW to manage nodes concurent access. This is why a logical drive per node is the best way ...
No ?

Thanks for this pedagogical exercice ...

[Doug de Werd · Accepted solution]

almost....

each drive is shared at the ACU/SSP level, but the cluster software gives ownership at the OS level. The cluster software cannot change ownership of the drive dynamically - meaning it takes at least a minute to switch it to the other server. There is really no simultaneous access to a drive.

Think of it like this. Server 1 owns drive F, and server 2 owns drive G. Under normal operations, the cluster software prevents server 1 from accessing drive G (and server 2 cannot access drive F). So it works just like a regular server. If you wanted server 1 to access drive G, you would have to fail over the disk resources (and other things) from server 2 to server 1. This would take about a minute, and it must be initiated manually. So now server 1 can access drives F and G, while server 2 cannot access either. If you want to give server 2 access to drive G again, you would have to reverse the process manually, and again it would take another minute.

If you want real simultaneous access of 2 servers to one LUN, you would need to use a product like PolyServe.

Here is some more information on PolyServe:

http://h71028.www7.hp.com/enterprise/cache/70995-0-0-225-121.html

http://h21007.www2.hp.com/dspp/mop/mop_PartnerDetails_IDX/1,2718,1489,00.html

http://www.polyserve.com/hp/index.php

Thanks,
Doug

[Iannuccelli]

Thanks for your help, I fell more confident now, waiting for the box delivery ...

[Iannuccelli]

My questions have been solved