Writes and deletes don't appear for a long time..... 1

[Prydonian]

OK. This is my situation…

I have 10 workstations (IBM ZPro Intellistations) and a Windows 2003 (IBM X335 eBlade) server networked together. All machines have 3.0 Ghz Xeon processors and 2.0 GB of memory. The workstations are diskless…. Client images of XP SP2 are stored on a Ciprico Talon II 560GB RAID 5.0 array. The server is running an application called Ardence Server BXP 3.3 which delivers the proper O/S image to the requesting workstation via Fast Ethernet (through an HP Procurve 4000M switch) depending on the client’s MAC address. The server also supplies IP addresses via DHCP. The workstations boot over the network using the PXE protocol. All of the workstations see this virtual image as their “C” drive. All this works perfectly. Performance is actually quite snappy.

The one problem I’m having is that a second RAID array in the Ciprico (also 560 GB) acts a data drive for all workstations and the server via fiber channel (QLA2200 HBAs) through a QLogic SANblade. All volumes are NTFS.

When a workstation (Any one. Doesn’t make a difference.) saves a file or folder, or for that matter, deletes a file or folder on the data RAID, the other workstations can’t see the changes for a long time period (sometimes hours….) while the workstation that performed the action can see and use it immediately.

According to Ciprico, write and delete caches are committed within millisecs and they haven’t heard of this problem before. The Ciprico has 512MB of cache on the RAID array…..

Anybody have any ideas or suggestions? I’m completely baffled!

Thanks in advance, Mike

 

[squiddog]

are you actually SHARING a volume via plain old FC connect?

Unless the hosts are using a shared disk via Windows Cluster (or someone elses clustering software), you are looking foward to a big disaster.

You cannot share a disk between hosts unless your hosts are clustered or running a cluster file system. NAS is the typical solution for shared files.

Don't use FC to share devices amongst non-clustered hosts. Each host thinks it has exclusive access to the disk.

Imagine that Host-A updates myfile.db. The updates are actually cached on Host-A. Host-B updates that file. Host-A doesn't see that. It goes to rewrite part of the file, but ends up corrupting it.



--
Bill Plein
a.k.a. squiddog
Contact me at

http://squadengine.com

 

[Prydonian]

Bill, you hit the nail on the head! Thanks for verifying I haven't lost my mind!

To fix this mess, I moved the machines in question off the Qlogic SANblade leaving the server that creates the data files via fiber channel and I'm sharing the data RAID over Ethernet. All machines are now happy and anotehr added benefit is that no longer are the workstations insisting on running CHKDSK on startup....

Thanks again, and a star for you!

Mike, The IT Guy.

 

[squiddog]

No problem.

Clustered File Systems are the Nirvana of SAN. A handful of people have done them or are working on them (active-active sharing of the same physical disk via block as opposed to a network file system like NFS or CIFS). The solutions for the most part are expensive and really only for specialized situations.

For most people, NAS is the right solution when trying to share files between servers.

--
Bill Plein
a.k.a. squiddog
Contact me at

http://squadengine.com