Shared storage considerations for DB2 pureScale environments

Fast I/O fencing

Before recovering a failed member in the DB2 pureScale instance, DB2 cluster services ensures that it is not possible for the failed member to modify shared data on disk. This precaution is known as I/O fencing. Some storage controllers and multipath I/O driver combinations support specialized SCSI commands, known as SCSI-3 Persistent Reserve (PR), with the write exclusive all registrants persistent reservation type. The SCSI-3 PR type write exclusive all registrants (type 7h) enables DB2 cluster services to perform fast I/O fencing, in as little as 1 - 2 seconds.

Recovery time is the elapsed time from a host failure until the release of locks for uncommitted transactions. For a typical OLTP workload with short transactions, fast I/O fencing results in recovery times of approximately 20 seconds. Fast I/O fencing is much faster than the alternative method, based on expiry of lock leases. When deciding which category to use, evaluate if fast I/O fencing support is required to meet your service level agreements (SLAs).

DB2 cluster services tiebreaker support

A subcluster of hosts requires operational quorum for IBM Reliable Scalable Cluster Technology (RSCT) to allow the subcluster to continue running the DB2 pureScale instance. To be granted operational quorum by RSCT, a subcluster must contain a majority of the hosts. If exactly half of the hosts in cluster remain in communication, the tiebreaker can be exclusively reserved by that subcluster to obtain operational quorum.

At the time of installation, the DB2 installer verifies whether a DB2 cluster services tiebreaker is specified and whether it can be used. Installation and instance creation is successful without a tiebreaker, however its absence results in a warning message that indicates a tiebreaker was not configured. Consider deploying an odd number of hosts to reduce the risk that failures leave a subcluster of half the original hosts.

On AIX® operating systems, if a tiebreaker fails to set during installation, you can manually set a tiebreaker by using the db2cluster command. Before you use the db2cluster command, you must run chdev on each host that will be a part of the DB2 pureScale cluster. The affected hosts must support the reserve_policy=single_path and algorithm=failover disk properties. You can set these disk properties by using the following commands:

chdev -l <disk_name> -a reserve_policy=single_path
chdev -l <disk_name> -a algorithm=failover

If you do not enable the reserve_policy=single_path and algorithm=failover disk properties, the tiebreaker might not be set or the data in the cluster might be corrupted.

On supported AIX operating systems, this tiebreaker disk is not a requirement because the tiebreaker reservation is using SCSI-2 reserve/release mechanism.

On supported Linux systems, the device used as the cluster manager (RSCT) tiebreaker disk must have SCSI-3 PR WRITE EXCLUSIVE REGISTRANTS ONLY reservation type code 0x50 enabled.

Subsystem Device Driver Path Control Module (SDDPCM)

Upgrading SDDPCM to 2.6.3.x might cause device configurations to fail if the algorithm attribute on the upgraded device is set to fail_over. If the device configuration settings fail, the DB2 pureScale cluster might be affected. You must set the timeout_policy attribute to retry_path if you want to use the fail_over option on the algorithm attribute. For more information, see http://www-01.ibm.com/support/docview.wss?&uid=ssg1S1004072.

Category 1 storage device and multipath I/O driver combinations

Storage devices and multipath I/O driver combinations listed in this category can successfully support both the DB2 cluster services tiebreaker and fast I/O fencing. Category 1 devices have been validated with the DB2 pureScale Feature and result in the highest resiliency and fastest recovery times.

Table 1. Category 1 storage device and multipath I/O driver combinations

Storage Devices	Multipath I/O drivers required for AIX systems	Multipath I/O drivers required for Linux systems	Protocol
IBM Storwize® V7000 (6.4.0.1 or higher)	SDDPCM	DM-MP	Fibre Channel
IBM SAN Volume Controller (6.4.0.1 or higher)	SDDPCM	DM-MP	Fibre Channel
IBM System Storage® DS8000® series	SDDPCM driver provided by IBM (driver file devices.fcp.disk.ibm.mpio.rte and devices.sddpcm<Your specific AIX version>.rte)	DM-MP	Fibre Channel
IBM System Storage DS5000 series	MPIO driver provided by IBM (driver file devices.fcp.disk.ibm.mpio.rte)	DM-MP or RDAC	Fibre Channel
IBM System Storage DS4000® series	MPIO driver provided by IBM (driver file devices.fcp.disk.ibm.mpio.rte)	DM-MP or RDAC	Fibre Channel
IBM System Storage DS3000 series	MPIO driver provided by IBM (driver file devices.fcp.disk.ibm.mpio.rte)	DM-MP or RDAC	Fibre Channel
EMC VMAX/Symmetrix family 1	MPIO driver provided by EMC (driver file EMC.Symmetrix.fcp.MPIO.rte)	DM-MP	Fibre Channel
NetApp FAS filers	MPIO driver provided by NetApp	DM-MP	iSCSI
Virtual I/O Server (VIOS) 2	MPIO or SDDPCM	Not Applicable	Fibre Channel
Hitachi Virtual Storage Platform (VSP)5	MPIO driver provided by IBM or HDLM	not supported as category 1 in Linux	Fibre Channel
Hitachi Universal Storage (HUS) 100 Series6	MPIO driver provided by IBM or HDLM	DM-MP	Fibre Channel
Hitachi Universal Storage (HUS) VM7	MPIO driver provided by IBM or HDLM	not supported as category 1 in Linux	Fibre Channel
Hitachi VSP G10007	MPIO driver provided by IBM or HDLM	not supported as category 1 in Linux	Fibre Channel
Note: Each host port connected to the EMC VMAX/Symmetrix family storage system must have the SCSI_3 (SC3) flag set on the front-end adapter port configuration. On the hypervolume, the logical unit number (LUN), used for the DB2 cluster services tiebreaker the SCSI3_persist_reserv option must be set. On supported Linux systems, the device selected for cluster manager (RSCT) tiebreaker, must support reservation type of WRITE EXCLUSIVE REGISTRANTS ONLY type code 5h. With SCSI-3 PR, there is one persistent reservation holder, which is the node that acquires the quorum device. A DB2 pureScale cluster using EMC VMAX shared storage, must have the SCSI3_persist_reserve type option enabled for the disks mapped. It is not enabled by default. A Virtual I/O Server must be used with N_Port ID Virtualization (NPIV) for a storage system and an AIX system multipath I/O driver combination. The VIOS storage is in the same category of storage that the underlying physical hardware storage system is in. For VIOS storage to be considered category 1 storage the physical storage hardware and driver combination must be in the category 1 table. Starting from SDDPCM v2630 and higher, before installing the DB2 pureScale Feature, you must change the timeout_policy attribute to retry_path on the tiebreaker disk. To change the value of the timeout_policy attribute, enter the following command: chdev -l hdiskX -a timeout_policy=retry_path For details, see: http://www-01.ibm.com/support/docview.wss?rs=540&context=ST52G7&dc=D600&uid=ssg1S1004072&loc=en_US&cs=utf-8&lang=en For more information about these drivers, see: For RDAC: http://www.lsi.com/sep/Pages/rdac/index.aspx For DM-MP on RHEL 5.5: http://docs.redhat.com/docs/en-US/Red_Hat_Enterprise_Linux/5/html/DM_Multipath/index.html For DM-MP on RHEL 6.1: http://docs.redhat.com/docs/en-US/Red_Hat_Enterprise_Linux/6/html/DM_Multipath/index.html For DM-MP on SUSE: http://www.novell.com/documentation/sles11/stor_admin/?page=/documentation/sles11/stor_admin/data/bookinfo.html The following restrictions apply: Supported only on AIX platforms HDLM (Hitachi Dynamic Link Manager) for AIX Version 7.3.1 or later provided by Hitachi Minimum micro code level 70-04-31-00/00 If HDLM is used, set the load balancing setting of the tiebreaker disk to OFF using HGLM (Hitachi Global Link Manager Software) Fast I/O fencing (SCSI-3PR) enabled Specific settings for disks that have fast I/O fencing enabled and are part of the GPFS file system: Host mode options 72 on Fibre channel only when HDLM is used The following restrictions apply: HDLM for AIX Version 7.3.1 or later provided by Hitachi Multipath I/O drivers required for Linux systems - DM- MP only Minimum micro code level 0945/A Select the Allocation length Expand mode check box if the disk is to be used as a tie breaker disk Fast I/O fencing (SCSI-3PR) enabled Select the Unique Reserve mode 2 check box for disks that are part of the GPFS file system and have fast I/O fencing enabled The following restrictions apply: Supported only on AIX platforms HDLM for AIX Version 7.3.1 or later provided by Hitachi HUS VM minimum micro code level 73-01-32-00/00 VSP G1000 minimum micro code level 80-01-41-00/00 If HDLM is used, set the load balancing setting of the tiebreaker disk to OFF using HGLM Fast I/O fencing (SCSI-3PR) enabled Specific settings for disks that have fast I/O fencing enabled and are part of the GPFS file system: Host mode options 72 on Fibre channel only when HDLM is used

Category 2 storage device and multipath I/O driver combinations

Category 3 storage device and multipath I/O driver combinations

All storage devices and multipath I/O driver combinations not identified as category 1 or category 2 fall into category 3. Category 3 combinations have not been validated with the DB2 pureScale Feature, though they are supported.

Recovery times for host failures in a DB2 pureScale environment are competitive with other vendors solutions, regardless of the category of storage device and multipath I/O driver.