|
HP OpenVMS Systemsask the wizard |
|
The Question is: Hi, I would appreciate any comments/hints you may have concerning the following configurations: Present configuration is a dual node AXP 8200/4100 SCSI cluster with a quorum disk. All disks are housed in an SW800 cabinet where we have 3 pairs of dual redundant HSZ50's serving approximately 350 Gigabytes of disk space configured using various RAID 5, 1,0 and 10 sets (24 logical VMS disks of sizes between 2 Gigabytes and 25 Gigabytes). Each node has 2 Gbyte memory and twin CPU's The application is a customer care and billing system using Ingres 6.4 which unfortunately is only supported on VMS upto 6.2-1H3 which is what we have. Our configuration of Ingres 6.4 does not know about clustering and so the 4100 runs the Ingres server processes whilst the 8200 takes the Ingres client processes and user logins (about 300 at present). The Ingres server <-> client traffic as well the SCA cluster traffic is handled by an isolated FDDI ring whilst user logins are by Telnet over 100Mbit ethernet. This seems to run fairly well at present, but the next stage is worrying me somewhat ... We are planning a disaster tolerant cluster by splitting the 2 nodes (and possibly upgrading the 4100 to the equivalent of an 8200) to different sites using a giga-sw itched FDDI ring for inter node comms and reproducing the disk configuration in another SW800 at the second site and then shadowing the whole lot using VMS volume shadowing. The quorum will be upheld using a third smaller node (eg a 2100) in a third site with a local system disk (locally shadowed for availability reasons). I have anticipated that flooding the FDDI ring during shadow copies is very likely and so plan to minimise their frequency by adding a medium sized extra node in each of the two main sites . These two extra nodes will do nothing much except ensure that th e shadow sets remain intact if and when one or both of the main application nodes goes down (lets call these smaller nodes the "shadow watchers"). To do this I figured on making all of these 4 nodes MSCP servers. I have two questions really ... at first I thought that I would have to assign cluster votes by giving 0 votes to each of the 2 shadow watchers and 1 vote to each of the two application nodes and 1 to the quorum watcher. This would ensure a cluster hang i f both application nodes crashed whilst theoretically still keeping both members of all shadow sets intact and so avoiding the need for expensive shadowset merges or copies when the nodes were brought back up. However, I have now also thought it may be better to keep the cluster up in such a circumstance by giving all nodes in the cluster exactly one vote. What worries me about doing this is what happens to the data in the database and the RMS flat files if bot h application nodes go down for some reason? Will the database of RMS flat files be corrupted? My second concern is for the length of time it will require to rebuild shadow sets amounting to 450 Gigabytes if and when they are required. Do you have any figures for this? I guess I also have a couple more questions... Will it work??? and have you seen a similar cluster elsewhere? Regards Anxious The Answer is : Please read and heed the information on VOTES and EXPECTED_VOTES that is present in the OpenVMS FAQ. RMS files will not be corrupted at loss of quorum, nor ensuing recovery. (The whole basis of the quorum scheme is to prevent these corruptions, by preventing partitioned clusters and thus preventing uncoordinated write operations to shared resources.) Shadow set merge operations occur when shadow set volumes are not cleanly dismounted -- this is not particularly tied to cluster quorum. Attempts to bypass the quorum mechanism can and have led to severe disk data corruptions. Discussions of disaster-tolerant configurations generally involves some assistance from folks very familiar with the local requirements, with the general pitfalls, and with the required final environment. (Disaster tolerance can initially look easy, but there are almost always a wide variety of considerations -- it is almost never as easy as it initially looks.) The OpenVMS Wizard would encourage you to contact the Compaq Customer Support Center for assistance and consulting services.
|