Re: [GNUnet-developers] [continued] insert - delete tests

[Christian Grothoff]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Thanks for the numbers Eric, I had planned to look into why gnunet-delete is 
so slow, but your numbers show that it's just the DB. About the fast download 
when the file is present -- GNUnet does not have to download blocks again if 
they are already there. If the entire file is present, gnunet-download just 
runs a check if it is really exactly the file and does not send a single 
request to gnunetd (!). It's essentially an extreme case of a download that 
is resumed. That's one of the really cool things about our encoding (and one 
of the reasons why the code also get's a big complicated).

Christian

On Friday 09 May 2003 08:59 am, eric haumant wrote:
> Hello,
>
> I've just run some tests on GNUnet 0.53 with a 100Mb file, BDB and MySQL.
> My tests was inserting the file (insertion of 106669 blocks), downloading
> it locally and deleting it several time.
>
> Here is the results with BerkeleyDB (I've done it just 10 times, it took
> too much time !!) :
>
> First time :
>  - insertion speed : 77 kb/s
>  - download speed : 850 kb/s
>  - deleting time : 34min (speed : 50 kb/s)
>  - place used :
>       before : 180 ko
>       after insert : 435Mo
>       after delete : 436 Mo
>
> Second time :
>  - insertion speed : 36 kb/s
>  - download speed : 113 kb/s
>  - deleting time : 1h (speed : 28 kb/s)
>  - place used :
>       before : 436 Mo
>       after insert : 437.9 Mo
>       after delete : 438.5 Mo
>
> ....
>
> 10th time :
>  - insertion speed : 36 kb/s
>  - download speed : 832 kb/s
>  - deleting time : 49min (speed : 39 kb/s)
>  - place used :
>       before : 438.6 Mo
>       after insert : 438.7 Mo
>       after delete : 438.7 Mo
>
> With BerkeleyDB the delete didn't reduce the database size but increase it
> a little with a maximum of 438.7Mo. After deletion we gain download speed.
> The BerkeleyDB is not really stable with test results, the deletion take
> from 29min to 4hours, not regular at all (strange). The insertion speed was
> though more regular : from 22min to 47min.
>
> Here is the results with MySQL (I've done it just 10 times, it took too
> much time !!) :
>
> First time :
>  - insertion speed : 51 kb/s
>  - download speed : 850 kb/s
>  - deleting time : 10min (speed : 170 kb/s)
>  - place used :
>       before : 150 ko
>       after insert : 115Mo
>       after delete : 115 Mo
>
> Second time :
>  - insertion speed : 49 kb/s
>  - download speed : 190 kb/s
>  - deleting time : 9min (speed : 190 kb/s)
>  - place used :
>       before : 115 Mo
>       after insert : 115 Mo
>       after delete : 115 Mo
>
> ....
>
> 10th time :
>  - insertion speed : 50 kb/s
>  - download speed : 190 kb/s
>  - deleting time : 10min (speed : 170 kb/s)
>  - place used :
>       before : 115 Mo
>       after insert : 115 Mo
>       after delete : 115 Mo
>
> With MySQL the delete didn't reduce the database size and like TrivialDB
> not increase it though. MySQL is the fastest for deletion and insertion.
>
> I've remarked that when I delete the file previously downloaded before
> calling gnunet-download, the speed was slower. I though gnunet didn't
> verify the file's content when downloading to an existing file. Is it due
> to disk access ? Is gnunet verifing file content if no IBlocks files is
> present ? For my test, if the file is present the download speeds up to
> 850ko/s but when not present it speeds up to 190ko/s(with mysql).
>
> Eric.
>
>
> _______________________________________________
> GNUnet-developers mailing list
> address@hidden
> http://mail.gnu.org/mailman/listinfo/gnunet-developers
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.0.6 (GNU/Linux)
Comment: For info see http://www.gnupg.org

iD8DBQE+u8sm9tNtMeXQLkIRAufOAKCr2pKKI2w1/8h3RWua67yNd1uYWwCeL3jJ
OwkcsUUIuUP7Zud4wOLNadg=
=Qc7r
-----END PGP SIGNATURE-----