January 9, 2003 at 11:35 pm
We have about 20 databases on our online SQL server, average size about 500M. These are data warehouse type of databases loaded once a month, but sometimes a few records would be corrected.
We would also like to have a mirror server in a branch, where we can import data and query locally, but replicate with the online server.
How would the performance be for 20 databases with only DSL connection? Will it be hard to manage? Should I use Transactional or Merge Repl?
Thanks,
James
January 10, 2003 at 3:49 am
May be a little slow on the initial process if done via Transactional replication simple because only changes are delivered there after. However, if something causes a need to resend the initial snapshot to sync again of course it gets slow again. Also, with Transactional replication you have a log monitor process running in memory for each DB that is being replicated so memory loads increase. TR is not hard to keep up with and with a slow connection or the need to keep up to the minute is better. Now with DSL a merge is better for the same condition as TR except that you are no trying to keep both synced on the minute just periodically. If you don't need instant gratification then merge is where I would go. But keep in mind it adds it's own columns for the merge to ID (I believe it adds a GUID which is several bits long) the record so it will add to the size.
January 10, 2003 at 8:45 am
It does add a guid and also restricts row length to 6k or so. Transactional is simple and fairly low impact. If you're running 20 log readers at once you'll burn maybe 40m of memory. Depending on the change rate you might get some contention in the distribution db. Changes will usually post quickly, probably in the 5-10 range for a single change, longer depending on transaction size.
Issue really is the snapshot if you're going to rebuild db (or bulk load more). I've run some over a 256k line, its slow but it works, and not slow enough to make me do something else.
I'd start with transactional for testing (because its simpler), get some times, then reassess.
Andy
January 13, 2003 at 12:22 pm
Thanks for answering my questions.
Hopefully, we'll do the snapshot replication only once, right? So that's not what I would concern. I tend to use Merge, and the GUID column doesn't matter to me much, if it doesn't affect my indexes size. And I don't reqire immediate sync.
I'm still a little confused about the resource consumption between Transactional and Merge Repl. Which will cost more memrory, CPU, and maintenance?
Thanks,
January 13, 2003 at 2:23 pm
I'd guess merge is more expensive because it adds trigggers everywhere. For transactional you just have the log reader running, 1-2m of memory, not a lot. Doesnt use much CPU. Maintenance - well, its replication. When it runs fine, no maintenance. When it breaks, well....
Andy
Viewing 5 posts - 1 through 4 (of 4 total)
You must be logged in to reply to this topic. Login to reply