Skip to Main Content

Hardware

How Do You Switch QFS Metadata Server Architectures

Here are the main points extracted from the SAM-QFS 4.4 release notes.

You cannot change the architecture type of the server responsible for control of  the file system metadata operations (i.e. the server that was used to create the file
system with the sammkfs(1M) command) by simply moving the disks from one server to the other. For a Sun StorEdge QFS standalone file system, this means that you cannot mount the file system on a server that has a  different architecture type from the one that created it. For a Sun StorEdge QFS shared file system, this means that you cannot change the architecture type of the metadata server or any potential metadata servers. This is because the different  architectures use different byte-ordering schemes (endianness). However, you can migrate data from one architecture type to the other by copying the file system to temporary storage using either qfsdump(1M) or samfsdump(1M), re-creating the file system using sammkfs(1M), and then repopulating the file system with qfsrestore(1M) or samfsrestore(1M).

EFI labels are required on all shared disks if your Sun StorEdge QFS shared file system configuration contains both the Solaris 10 OS on x64 platforms and the Solaris 9 or Solaris 10 OS on SPARC platforms.

Use the Solaris prtvtoc(1M) command to determine whether a disk contains SMI or EFI labels. Under the Dimensions section of the output, SMI labels list the number of accessible cylinders, whereas EFI labels list the number of accessible sectors.

To convert disk labels from the default SMI VTOC8 to EFI, copy the file system to temporary storage using qfsdump(1M) or samfsdump(1M), relabel the disks with EFI labels using the Solaris format -e command, re-create the file system using sammkfs(1M), and repopulate the file system with qfsrestore(1M) or samfsrestore(1M).

When using the Solaris format -e command to create EFI labels, you can select the partition command from the menu to create and modify partitions (slices). When doing this, you must specify a tag id name of usr, rather than stand or unassigned, for EFI labels.

Note that EFI labels reserve the first 34 sectors, which misaligns Sun RAID-5 storage from a performance perspective. Unless you realign the storage, you will incur a RAID-5 read/modify/write performance penalty when writing. You can avoid this performance penalty by selecting the proper starting sector for all disk partitions for your particular storage configuration. For example, an 8+P Sun StorEdge T3 array with a 64K block size should have starting sectors that are multiples of 1024 for all disk slices ((8 * 64 * 1024) / 512 = 1024). Similarly, a 5+P Sun StorEdge FC 3510 array with a 128K block size should have starting sectors that are multiples of 1280 for all disk slices ((5 * 128 * 1024) / 512 = 1280).

Comments
Post Details