Re: Cached ADO
- From: "Anwar Shafiev" <AnwarShafiev@xxxxxxxxxxxxxxxxxxxxxxxxx>
- Date: Thu, 30 Jun 2005 11:49:08 -0700
Stephen, thanks for the information. This is a kind of experience, which we
cannot get just reading books. :) I agree with you. Any solution must be
reasonable. In my case, I used a main map (primary key, cached object) and a
vector of multimaps for indexes. As a result, searching on indexed fields
takes some nanoseconds. I use static binding to work with recordsets. I wrote
a wizard (btw, that's what I miss too in classic ADO) to generate both a
static binding and a cached object classes out of database structures. So
working with the layer is pretty easy. The cached recordset class is
impemented in a template. API to the cached recordset is similar to ADO. I
don't want to go deep into details, but I am satisfied with the performance,
although the functionality is very limited. Well, I am almost satisfied. I
think, any programmer will never be satisfied with the results. :) The
perfomance is acceptable, but I really feel a lack of functionality. I cannot
believe that nothing similar exists. At least, I know many developers who are
forced to create their own caching layers like me. :-/
Anwar
"Stephen Howe" wrote:
> > Stephen, my caching layer is based on STL too. I use maps, as they provide
> > much faster indexed searching on a large emount of data.
>
> I also use maps but there are some riders to this:
> (i) I only use maps if I have true key-value pairs that are logically
> distinct. If I am dealing with a single entity (or class) then I use a set.
> It does not make sense to break a class apart or repeat the key element
> _just_ to use a map. It took me a long time (months) to realise this.
> (ii) For 2000 items or less, a vector will beat a map. Every time you insert
> into a map, that is a memory allocation. That happens less often for vector.
> And if you know the size in advance vectors reserve() will cut down on
> memory reallocation. The fact that vectors insert is O(N) and maps is O(log
> N) barely matters if copying objects is cheap. map only really comes into
> its own when copying is not cheap and there is a large number of items. You
> have to have a very large number of items before map's O(log N) beats
> vectors O(N) due to the memory allocation/deallocation acting as a drag to
> map's performance .
> That 2000 items could be 5000 for say integers and less for heavy-duty
> classes.
> (iii) If a container once populated rarely changes, a sorted vector searched
> using lower_bound() is far faster than maps find. If it changes once in a
> blue moon, sorted vector vector still wins.
>
> This comes down to : Roughly I use map/set when > 2000 items and number of
> items is changing constantly over lifetime of container. Iterator validity
> is also a consideration.
>
> > As I already
> > described in my previous post to Bill, the basic functionality exists. But
> > that's not enough for me. I need to have powerful filtering capabilities,
> > extremely fast searching and data manipulation. The layer must be
> thread-safe
> > and must solve all collisions in multi-user environment. I would like to
> have
> > transactions, referential integrity, relationships between tables on the
> > caching layer too.
>
> Tough requirements. I don't see how "collisions" are solved intelligently.
> Problem is your describing a seemless world - which we do not have.
>
> In C++, you can write thread-safe code but it is not an automatic given.
> The programmer usually has to supply some design intelligence to make it
> thread-safe and of course bugs can creep in. I hate that. There is no way a
> newbie C++ programmer can _easily_ write thread-safe code. It requires an
> experienced programmer apply intelligence. And C++, as a language, does not
> make guarantees on thread-safety. A real problem.
>
> In the database world, SQL Server and some of Microsoft's client-sided ADO
> supplies searching, filtering, transactions, referential integrity,
> relationships but these relationships are not maintained once you have the
> data inside the client app.
>
> So it is a mish-mash. You ruled out .NET but I believe it makes some inroads
> in presenting a seemless environment. I don't think you will get a seemless
> environment with C++ and classic ADO.
>
> >Well, now when I think about all these things, which I
> > have to implement, I decide to buy an existing solution if it exists, of
> > course. :)
>
> Sure :-)
>
> >If not... Then I will continue developing my own layer.
>
> It sounds very similar to what we have :-)
> Of late I am using a class to represent a table and it is the class's
> responsibility to perform any read/updates to the table. The class caches
> the table as it see fit. All other parts of the app, deal with the class and
> its interface.
>
> Stephen Howe
>
>
>
>
>
>
>
>
.
- References:
- Cached ADO
- From: Anwar Shafiev
- Re: Cached ADO
- From: Stephen Howe
- Re: Cached ADO
- From: Anwar Shafiev
- Re: Cached ADO
- From: Stephen Howe
- Cached ADO
- Prev by Date: Re: Prompting for user id/password when using Integrated Security
- Next by Date: Re: Finding row by a non-primary key
- Previous by thread: Re: Cached ADO
- Next by thread: Finding row by a non-primary key
- Index(es):
Relevant Pages
|
