Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.misc -> Replication; Snapshots; Refresh questions...

Replication; Snapshots; Refresh questions...

From: <cew120463_at_my-dejanews.com>
Date: Wed, 26 Aug 1998 17:28:08 GMT
Message-ID: <6s1gj8$m41$1@nnrp1.dejanews.com>

Our architecture is Single-Master (OE8 on RS6000), Multi-Snapshot (PO8 on Thinkpads). Laptops will generally refresh by modem, sometimes via lan.

We have a read-only snapshot group (of tables) that is being updated in a very "Sparse-Heavy" fashion at the master site (never modified at laptop). The scenario is "inventory". The inventory tables are updated at the server every 5 minutes. In the course of 24 hours, "hot" items will see a _LOT_ of updates (same "row" changes virtually every 5 minutes), but the bulk of the inventory will remain unchanged. If the snapshot is defined as FAST-REFRESH, will Oracle be forced to store _every_ update in the snapshot log, or can we tell it to "discard" previous updates from the log? Each laptop _only_ cares about the very latest values. All previous values are irrelevant.

I had always thought that since it was a "Read-Only" snapshot, that the server could just maintain the latest values and "discard" previous updates from the log. However, after thinking about it, I suspect that for "transactional consistency" type reasons, there may be no way for Oracle to flush the log until _every_ snapshot has received _every_ individual (5-minute interval) update. If a laptop doesn't synch for a couple of days (or worse-yet... a week), the log could very likely become immense!

Perhaps such a snapshot is better suited to COMPLETE-REFRESH, but this is undesireable since laptops are expected to dial-in for current inventory regularly. It's the "guy who goes on vacation" scenario which ends up dictating the replication scheme. Are we forced to "protect against the exception" as opposed to "optimizing for the rule"? Please tell me a way for read-only snapshots to ignore the revision-history and just receive the latest value...

Is there a formula to estimate worst-case log file requirements?

Is there a rule (aside from trial&error testing/tuning) to use when deciding between FAST vs COMPLETE refreshes to snapshots?

Are there any decent books out there yet on replication?

Huge Thanks in Advance for any info you can share...

Chuck

-----== Posted via Deja News, The Leader in Internet Discussion ==----- http://www.dejanews.com/rg_mkgrp.xp Create Your Own Free Member Forum Received on Wed Aug 26 1998 - 12:28:08 CDT

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US