Re: IDS 11.5 and SDS

Tom wrote: > Greetings All, > > Has anyone configured IDS 11.5 with SDS on any AIX platforms? I am > configuring two new servers and plan on using IDS 11.5 features for HA > and it looks like a Shared Disk Secondary would fit the bill. However, > a call to IBM Informix as well as IBM AIX Software group provided no > info on how to configure AIX using raw LV's to meet the "Shared" disk > requirement for SDS to work. > > The two servers are identical P6-550's connected via SAN to an IBM > 2105-800. The IDS is 11.5.FC3. > > Thanks, > > Tom Tom, The only real requirement is that the shared disk subsystem ensure that when the server issues a write IO that the write makes it all the way to the physical disk before returning a write complete status. We can not support any form of shared disk environment in which the write remains in a buffer - be it OS or disk cache - and then is trickle written to the disk. There is a bit of coordination between the disk writes and the SDS nodes which must be supported. The information that you are getting from tech support is based on what we have tested with, but we do have some customers who have either tested or are in production with shared disk solutions outside of what we tested with internally. M.P.

Madison Pruet wrote: > Tom wrote: >> Greetings All, >> >> Has anyone configured IDS 11.5 with SDS on any AIX platforms? I am >> configuring two new servers and plan on using IDS 11.5 features for HA >> and it looks like a Shared Disk Secondary would fit the bill. However, >> a call to IBM Informix as well as IBM AIX Software group provided no >> info on how to configure AIX using raw LV's to meet the "Shared" disk >> requirement for SDS to work. >> >> The two servers are identical P6-550's connected via SAN to an IBM >> 2105-800. The IDS is 11.5.FC3. >> >> Thanks, >> >> Tom > > Tom, > > The only real requirement is that the shared disk subsystem ensure that > when the server issues a write IO that the write makes it all the way to > the physical disk before returning a write complete status. and that the read comes from the physical disk and not from OS/disk memory buffers. In other words, if node A writes a page and node B subsequently reads that same page, then node B must read what A wrote... We can not > support any form of shared disk environment in which the write remains > in a buffer - be it OS or disk cache - and then is trickle written to > the disk. There is a bit of coordination between the disk writes and > the SDS nodes which must be supported. > > The information that you are getting from tech support is based on what > we have tested with, but we do have some customers who have either > tested or are in production with shared disk solutions outside of what > we tested with internally. > > M.P.