Hello Fabrice
Given:
<items>
<item><ID>[a number]</ID><Type>[a string]</Type></item>
…
<item><ID>[a number]</ID><Type>[a string]</Type></item>
</items>
The size is ~5m `item`. (Depending on the query, we are talking about a few
million items)
If I don’t add any external additional structure, which here is defined by the
`item`, `items` elements, then the “unformatted” output is generated in under
2sec.
<ID>[a number]</ID><Type>[a string]</Type>
…
<ID>[a number]</ID><Type>[a string]</Type>
Again, that would be a few million items. Queries are exactly the same apart
from the addition of element items{ for… for… return element item {…}}
The problem with this second representation is that you don’t really know where
tags from one item of the original database begin and end, this is why I want
to enclose them further.
All the best
All the best
From: basex-talk-boun...@mailman.uni-konstanz.de
[mailto:basex-talk-boun...@mailman.uni-konstanz.de] On Behalf Of Fabrice
ETANCHAUD
Sent: 18 September 2017 15:56
To: basex-talk@mailman.uni-konstanz.de
Subject: Re: [basex-talk] Basex Inner Workings
Hi Athanasios,
Could you please give us a idea of your resulting document size after 1.5
minutes of BaseX time ?
Best regards,
Fabrice
De :
basex-talk-boun...@mailman.uni-konstanz.de<mailto:basex-talk-boun...@mailman.uni-konstanz.de>
[mailto:basex-talk-boun...@mailman.uni-konstanz.de] De la part de Anastasiou A.
Envoyé : lundi 18 septembre 2017 14:47
À : 'Graydon Saunders';
basex-talk@mailman.uni-konstanz.de<mailto:basex-talk@mailman.uni-konstanz.de>
Objet : Re: [basex-talk] Basex Inner Workings
Hello
Many thanks, Dirk, Fabrice and Graydon.
I was going to look up ways of enabling the server to run as fast as possible
anyway later on, so it is always good to know how is BaseX “thinking”.
I can see what you mean Graydon. This is a simple nested `for` to denormalise
some of the structures of the XML file, where “some” is defined by
an XPath expression.
As far as I can tell, there is nothing being re-evaluated repeatedly within the
inner loop that could be brought outside.
I have gone through the dot plans of the quickest and slowest versions of the
query and the only thing they differ is in the addition of the CElems.
The “scaling” of the timings, in case it helps, is as follows:
Simple query, returning elements: 1100-1500 ms
Adding an `element` to what is returned just by the innermost `for`: 7500-9311
ms
This means:
For…
For….
Return element item{someElement|someOtherElement}
Adding an `element` to the whole block (no `element` to the innermost
`for`):49000-67000ms
This means:
Element Items{
For…
For…
Return someElement|someOtherElement
}
Adding an `element` to both places: 50000-80000ms
This means:
Element Items{
For…
For …
Return element Item {someElement|someOtherElement}
}
I don’t mind the ~8sec time but when we get to 1.5min, then yes…that’s going to
be a bit annoying.
All the best
From: Graydon Saunders [mailto:graydon...@gmail.com]
Sent: 15 September 2017 17:04
To: Anastasiou A.;
basex-talk@mailman.uni-konstanz.de<mailto:basex-talk@mailman.uni-konstanz.de>
Subject: Re: [basex-talk] Basex Inner Workings
As a follow-on to Dirk, it's amazing how much of a performance difference it
can make to use typed variables when you're constructing something for output.
(So far as I can tell, variables declarations function as an "optimize this!"
flag for BaseX.)
If you get good performance when you're just throwing the resulting nodes and
lose it massively by adding structure, as you relate up there somewhere are:
The change was to go from simply returning the nodes themselves with a `return
thisnode | thatnode |theothernode` to a "formatted" document that has an outer
<collection> with a number of `return
<item>{thisNode|thatNode|theOtherNode}</item>` inside it.
my immediate thought was "it's querying the same thing multiple times".
Most programming languages it's good practice to not create variables when you
can inline. XQuery does not appear to be one of those languages. :) I try to
think of this as "how can I make things easy for the optimizer?"
-- Graydon
On Fri, Sep 15, 2017 at 11:55 AM, Kirsten, Dirk
<dirk.kirs...@senacor.com<mailto:dirk.kirs...@senacor.com>> wrote:
Hello Athanasios,
I think you should really check the actual query plan which is executed. If you
have such a huge spike in performance surely they processor will be executing
it differently. I don't think looking into file access patterns BaseX
internally uses is very useful for an end user. You should let BaseX handle
that (but of course, if you find better/more efficient ways I am sure
Christian' gladly accepts Pull Requests). But the pattern you describe sounds
very much excepted, so reads if you open databases seem logical and short write
operations are also expected when just reading a database, because e.g. BaseX
has to lock the databases.
So I think it would be more useful to look into the query plan. Of course you
are more than welcome to ask about what is going on there on this list. I would
expect that because of your rewrite maybe some indexes are not applied anymore
(or if your rewrite is simply very big that most of the time is spent
serializing the data).
Cheers
Dirk
Senacor Technologies Aktiengesellschaft - Sitz: Eschborn - Amtsgericht
Frankfurt am Main - Reg.-Nr.: HRB 105546
Vorstand: Matthias Tomann, Marcus Purzer - Aufsichtsratsvorsitzender: Daniel
Grözinger
-----Ursprüngliche Nachricht-----
Von:
basex-talk-boun...@mailman.uni-konstanz.de<mailto:basex-talk-boun...@mailman.uni-konstanz.de>
[mailto:basex-talk-boun...@mailman.uni-konstanz.de] Im Auftrag von Fabrice
ETANCHAUD
Gesendet: Freitag, 15. September 2017 17:35
An: 'Anastasiou A.'
<a.anastas...@swansea.ac.uk<mailto:a.anastas...@swansea.ac.uk>>;
basex-talk@mailman.uni-konstanz.de<mailto:basex-talk@mailman.uni-konstanz.de>
Betreff: Re: [basex-talk] Basex Inner Workings
You can find the time spent in each step in the query info bar graph.
If you are looking for the schema and the facets of your dataset, you should
have a look at the index module, and for sure at index:facets()
Best regards,
Fabrice
-----Message d'origine-----
De : Anastasiou A. [mailto:a.anastas...@swansea.ac.uk]
Envoyé : vendredi 15 septembre 2017 17:23 À : Fabrice ETANCHAUD;
basex-talk@mailman.uni-konstanz.de<mailto:basex-talk@mailman.uni-konstanz.de>
Objet : RE: Basex Inner Workings
Thank you Fabrice. I understand.
I have not tried querying from the command prompt or sending the output to a
file directly, which I could also work with. But, my understanding is that the
time we are being quoted by the gui is the DB time, not taking into account the
time it takes for the list to be pushed into whatever data structures the list
boxes might be supporting (?).
I am trying to get a better understanding of the dataset at the moment and I
have short and long queries which depending on the results I get from this step
could be optimised further.
All the best
-----Original Message-----
From: Fabrice ETANCHAUD [mailto:fetanch...@pch.cerfrance.fr]
Sent: 15 September 2017 16:17
To: Anastasiou A.;
basex-talk@mailman.uni-konstanz.de<mailto:basex-talk@mailman.uni-konstanz.de>
Subject: RE: Basex Inner Workings
I understand that you are reformatting a lot of data, aren't you ?
I will have only little advice, because this is not my use case.
From what I know, resulting document will be materialized entirely in memory
before presentation or export.
You should export your results to disk, in order not to lose time in BaseXGUI
rendering.
To reformat very big amounts of data, you might have a look at saxon streaming
features (not in the free version).
But usually, big results are not requested frequently.
Best regards,
Fabrice
-----Message d'origine-----
De : Anastasiou A. [mailto:a.anastas...@swansea.ac.uk]
Envoyé : vendredi 15 septembre 2017 16:39 À : Fabrice ETANCHAUD;
basex-talk@mailman.uni-konstanz.de<mailto:basex-talk@mailman.uni-konstanz.de>
Objet : RE: Basex Inner Workings
Hello Fabrice
Yes, I am having a query which jumped from ~1500 ms to about a minute with a
tiny little change...
The DB is about 2GB and it is my test set before putting the query to work on
the full dataset.
The change was to go from simply returning the nodes themselves with a `return
thisnode | thatnode |theothernode` to a "formatted" document that has an outer
<collection> with a number of `return
<item>{thisNode|thatNode|theOtherNode}</item>` inside it.
I understand that the new query might be creating some new entities but
compared to the element content, these few extra characters are not THAT many
more.
The query jumps from ~1500 ms when using plain XML, to ~55000ms with the
addition of the collection, item nodes, to ~57000ms with the addition of CSV
exporting via the CSV module. These are "informal average" values. So, I have
not run the same query a few times and then obtain the average, but that's the
sort of vicinity I have seen numbers in from the times I have run the queries
so far.
The database itself is "static", there are no update/insert transactions at the
moment, the only thing that I am trying to do is extract some data in a
different format from it.
I have Text, Attribute and Token indexes on that database (optimised right
after importing) but no further options enabled. I also have not experimented
with the SPLITSIZE (?). I have 32GB of memory and it should be enough to handle
this 2GB test dataset (?). I will have a go with DEBUG on.
Did you have to enable any additional options for indexes to work faster?
All the best
-----Original Message-----
From:
basex-talk-boun...@mailman.uni-konstanz.de<mailto:basex-talk-boun...@mailman.uni-konstanz.de>
[mailto:basex-talk-boun...@mailman.uni-konstanz.de] On Behalf Of Fabrice
ETANCHAUD
Sent: 15 September 2017 13:27
To:
basex-talk@mailman.uni-konstanz.de<mailto:basex-talk@mailman.uni-konstanz.de>
Subject: Re: [basex-talk] Basex Inner Workings
Hi Athanasios,
Did you experience slow queries ?
Are you sure to use all the index features ?
Are these queries operational ones (direct access on a key value) or analytics ?
I never experienced slow queries, even on huge xml corpus (patent
registrations), But this is at the cost of longer indexing times on updates.
Best regards,
-----Message d'origine-----
De :
basex-talk-boun...@mailman.uni-konstanz.de<mailto:basex-talk-boun...@mailman.uni-konstanz.de>
[mailto:basex-talk-boun...@mailman.uni-konstanz.de] De la part de Anastasiou A.
Envoyé : vendredi 15 septembre 2017 14:01 À :
basex-talk@mailman.uni-konstanz.de<mailto:basex-talk@mailman.uni-konstanz.de>
Objet : [basex-talk] Basex Inner Workings
Hello everyone
Quick question: Is there any document / URL where I could find out more about
how does Basex access the disk during its operation?
For example, are there any reads to be expected during executing a query?
Through iotop, I can see 3-4 processes reading during startup, then another 2,
very briefly firing when opening the database and then during querying there
are periodic writes (?) but of very brief duration.
I was wondering if there is anything that could be done from the point of view
of the hardware to speed up queries (?) (except a more powerful machine at the
moment)
All the best
Athanasios Anastasiou
