Message-ID: <4C2308FD.1090300@cybertec.at>
Date: Thu, 24 Jun 2010 09:27:57 +0200
From: =?ISO-8859-1?Q?B=F6sz=F6rm=E9nyi_Zolt=E1n?= <zb@cybertec.at>
User-Agent: Mozilla/5.0 (X11; U; Linux x86_64; en-US;
	rv:1.9.1.10) Gecko/20100621 Fedora/3.0.5-1.fc13 Thunderbird/3.0.5
MIME-Version: 1.0
To: Bruce Momjian <bruce@momjian.us>
CC: Michael Meskes <meskes@postgresql.org>,
	PG Hackers <pgsql-hackers@postgresql.org>,
	Hans-Juergen Schoenig <hs@cybertec.at>
Subject: Re: ECPG FETCH readahead
References: <201006232042.o5NKgb503695@momjian.us>
In-Reply-To: <201006232042.o5NKgb503695@momjian.us>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 8bit

Hi,

2010-06-23 22:42 keltezéssel, Bruce Momjian írta:
> Boszormenyi Zoltan wrote:
>    
>> Hi,
>>
>> we improved ECPG quite a lot in 9.0 because we worked and
>> still working with an Informix to PostgreSQL migration project.
>>
>> We came across a pretty big performance problem that can be seen in
>> every "naive" application that uses only FETCH 1, FETCH RELATIVE
>> or FETCH ABSOLUTE. These are almost the only FETCH variations
>> usable in Informix, i.e. it doesn't have the grammar for fetching N rows
>> at once. Instead, the Client SDK libraries do caching themselves
>> behind the scenes to reduce network turnaround time.
>>      
> I assume our ecpg version supports>1 fetch values, even in Informix
> mode.  Does it make sense to add lots of code to our ecpg then?
>    

I think, yes, it does make sense. Because we are talking
about porting a whole lot of COBOL applications.
The ESQL/C or ECPG connector was already written
the Informix quirks in mind, so it fetches only one record
at a time passing it to the application. And similar performance
is expected from ECPG - which excpectation is not fulfilled
currently because libecpg doesn't do the same caching as
ESQL/C does.

And FYI, I haven't added a whole lot of code, most of the
code changes in the patch is execute.c refactoring.
ECPGdo() was split into several functions, the new parts
are still doing the same things.

I can make the test case much smaller, I only needed
to test crossing the readahead window boundary.
This would also make the patch much smaller.

And this readahead is not on by default, it's only activated
by "ecpg -r fetch_readahead".

Best regards,
Zoltán Böszörményi