Re: IBM Debunks Oracle's MultiVersion Read Consistency ?

From: Dave Hau <davehau-no-spam-123_at_no-spam.netscape.net>
Date: Sun, 20 Apr 2003 14:08:02 -0700
Message-ID: <b7v12q$2u4$1@slb9.atl.mindspring.net>

"Robert Allen" <rallen_at_NOSPAMOKhgi.com> wrote in message news:b_yoa.380076$HU.52078_at_news.easynews.com...
> http://www-
> 3.ibm.com/software/data/pubs/papers/readconsistency/readconsistency.pdf
>
> They raise some good points. Is it really as efficient as IBM suggests?
> What are the advantages that MVRC provide?

A very significant advantage of MVRC not mentioned by the paper is the reduced risk of deadlocks in an OLTP application, because MVRC means you don't need to hold a read lock to read a row. As any OLTP app developer would agree, deadlocks are hard to avoid unless you adopt a rigorous scheme to access resources in a fixed order. Therefore, the fewer locks you need to hold in a transaction, the lower the risk of deadlocks. I see this as a huge advantage of MVRC in a highly concurrent OLTP system.

Once you hit a deadlock, you either have to wait till your transaction times out, or wait for the database to detect the deadlock and kill off one of the threads. Most databases can detect 2-way deadlocks but in the (rare but existing) case of 3 or more way deadlocks, your thread will be sitting there until timeout (i.e. if timeout is implemented in the app).

Cheers,
Dave Received on Sun Apr 20 2003 - 16:08:02 CDT