IBM TDS/SDS (LDAP) - can I mix endianness among servers in an instance ?

I'd like to add some x/linux-based servers to my current AIX-based TDS/SDS server community. Reading the Fine Install Guide (rtfig ?) I believe this may be covered by the section "Upgrade an instance of a previous version to a different computer" i.e. I'm going to install latest/greatest SDS on a new server image, then bring in the instance data from one of my current servers.

The chart in that section of the Install Guide indicates this is only possible among systems with the same endianness.

Is it possible to have Intel & power servers cooperating on the same directory ? If so, can someone pls point me to doc on how to get there ?

Much appreciated !

[edit:] more reading. I see in the section "creating a copy of an existing instance" the same requirement that only data from a source with the same endianness as the target can be used.

So: can I create an instance copy without data, then (quiesce replication and suspend my servers and) dump the directory to an LDIF, and load the new instance from that instead of from a backup ? Then ensuring that replication is prepared on the new copy and then spinning up the whole pile ? The crux being that by importing the data as text rather than binary, I avoid the fact that handling of endianness/network byte order is missing from the deploy tools ?

Same question as above though, crudely, does replication handle network byte order correctly and swap data nice between different endiannesses ?

Why not just set up a test bed and give it a try?

That's the easiest way (when in doubt and the docs are not clear). Set up a non-production test bed and try it out.

:slight_smile: yep. I figured if someone has blazed this trail before I'd sponge off their work. I'll update here with results.

well endianness only matters for binary data not text data. If I understand you correctly you want to transfer text data only so that should work fine. If you however need something runnable you would need to run it through a converter and recompile.

1 Like

ibm sez yes it's supported. we'll give it a whirl. will update with highlights.

--- Post updated at 11:16 ---

ibm suppt said that replication does the same ops on the recipient that were done on the originator to effect the change in the first place. which says to me all the data sent in replication must be text, I'll check that with tcpdump when I get time. I suspect it falls down on binary blobs like image data/video saved in attributes, but we don't have any of those so we don't care. Passwords might be fun too but OTOH those have their own handling including cryptosync, so I expect they're correct.

In general, my reading (is it correct ?) of this is that it's about correct layer6 implementation, and Sun took care of that over 30 years ago with XDR for RPC, later standardized into the notion of network byte order.
that was cool, I made my own editorial tags and whaddya know, someone had implemented them. who knew... :stuck_out_tongue:

Moderator comments were removed during original forum migration.