On 14 May 2018 at 01:08, Richard Damon <rich...@damon-family.org> wrote:

> On 5/13/18 12:55 PM, Rowan Worth wrote:
> > On 9 May 2018 at 08:56, Richard Hipp <d...@sqlite.org> wrote:
> >
> >>   But with
> >> SQLite, there is no round-trip latency.  A "round-trip" to and
> >> database is just a function call, and is very very cheap.
> >>
> > I want to emphasise that Dr. Hipp's usage of "round-trip" only includes
> the
> > latency of _communication_ between the app and database in this
> statement,
> > and excludes any processing time required by the database.
> >
> > If you were to interpret "round-trip" from an app-centric perspective (as
> > in "the time taken to retrieve/commit data") then the statement becomes
> > misleading because handling the data involves i/o, possibly even
> > synchronous i/o, which is not "very very cheap" by any standard I'm aware
> > of :)
> >
> > -Rowan
>
Yes, if the request requires I/O, then that costs time, but then the
> operation would likely use similar I/O in whatever way the application
> needed to get that information, so that I/O shouldn't really be charged
> to the use of a database, but to the information requested. One thing to
> remember is SQLite is often compared as a better way to get information
> then using simple disk i/o, so the 'cost' of using SQLite (compared to
> the alternative) shouldn't include the base time to read the file, but
> only any extra i/o above that.
>

That's fair, but I wouldn't call the extra i/o imposed by sqlite "very very
cheap" either - it doubles writes (once to the rollback journal, once to
the DB), forces syncs, and likely results in a more seek heavy i/o pattern
(this depends a bit on schema design and whether the app requires/fully
takes advantage of relational operations).

To be clear, this is not a criticism of sqlite. These costs are paid for a
reason (eg. durability) and I think sqlite does its job very efficiently.
You're also right that an app implementing similar features without sqlite
will have to pay similar costs.

My point is simply that it's unwise to think of any DB query as having "no
latency" even when dealing with an sqlite DB.
-Rowan
_______________________________________________
sqlite-users mailing list
sqlite-users@mailinglists.sqlite.org
http://mailinglists.sqlite.org/cgi-bin/mailman/listinfo/sqlite-users

Reply via email to