Subject: Re: Sherlock from Mac OS 9 slow
From: Lorenzo Perone (lopez.on.the.lists@yellowspace.net)
Date: Thu Mar 29 2001 - 20:18:18 EST
At 17:19 Uhr -0500 29.03.2001, Vic Landi wrote:
>Problem is that _find_ is usually generated every night by cron... The mac
>does it on the fly... if you wanted to repeat the process, you'd be choking
>up a lot of time running the update every second...
You seem to be right. /usr/bin/finding on 60gigs of hds took 10 mins
14 secs.. (G3 DT/ 266)
while OS9 on a G4 needs _20 secs_ on the same amount of data. It
looks like some kind of index is being made 'on the fly'...
But: the same search that took 46 secs with find -iname, took _25_
minutes with netatalk to complete (for 625 result items in 16 Gigs).
that can't be true.
Find is terribly slow but, still, usable - waiting for one
neta-search to complete you get old.
It would still be an improvement x30 to implement find's methods...
on the other hand, as for alternatives: i think a combination of
apache (Options Indexes et al.), htdig and a sherlock-plug-in might
do a good job for content-based finding, that could partially ease
the pain of not having decent name searches. I still don't know what
Sherlock does with a link like file:///folder/file.... he might as
well try to open it with a browser and for sure there'll be no 'Open
Enclosed Folder' option available. So it might also involve writing a
small 'Neta-htdig-Sherlock....' or an app handling a 'special
protocol' like netatalk://. In this case one could search with a
browser and the link would then reveal (via a special app catching
the appropriate event) the file.... yes, it's getting tricky.
;-!
Lorenzo
>Don't ask me about the file system thing... I'm just relating what I was
>told the last time when I brought this up...
>
>> From: Lorenzo Perone <lopez.on.the.lists@yellowspace.net>
>> Date: Thu, 29 Mar 2001 23:04:55 +0200
>> To: Vic Landi <vic@mail.jamesltaylor.com>
>> Cc: <netatalk-admins@umich.edu>
>> Subject: Re: Sherlock from Mac OS 9 slow
>> Resent-From: netatalk-admins@umich.edu
>> Resent-Date: Thu, 29 Mar 2001 16:12:35 -0500 (EST)
>>
>> At 15:32 Uhr -0500 29.03.2001, Vic Landi wrote:
>>> I've been looking for a solution as well, however I'm afraid there is not
>>> one yet. The problem is that the Mac OS uses a db to make the searches
>>> faster (a hidden file that makes records on where everything is) So when
>>> you do a sherlock search on a netatalk server you actively search through
>>> the directory tree, taking a lot of time. Solutions to this have been
>>> posed... These include the ext3 file system (if and when it comes out) and
>>> rasierfs (this I'm unsure of)
>>
>>
>> Hi!
>>
>> I don't understand why a different filesystem should be needed.
>> I mean, _find_ works way faster than sherlock searches on netatalk
>>- wouldn't
>> it be possible to link to find or its library, or use its sources
>>to implement
>> a faster search?
>> I made some tests over a Perl CGI using find (plannig to make a sherlock
>> plug-in with file:// urls), and though slower than sherlock-on-asip, it was
>> very usable for what concerns the speed (on netatalk sherlock is unusable,
>> let's face it - at least on volumes larger than 4GB).
>> Unfortunately, because my netatalk is not yet using codepages, I can't use
>> this CGI yet (url / character conversion issues...).
>> If I get to finish it and it looks usable I can post that cgi here, but I'd
>> prefer somebody could implement something like find-in-netatalk......
>>
>> :-)
>>
>>
>> Lorenzo
>>
>>
>>> However, from what I saw in the public beta of OS X, even that was slower,
>>> but not as slow... could it be that apple has devised a way of their own?
>>>
>>> Vic Landi
>>>
>>>> From: "Bernard M. Piller" <bernard@bmpsystems.com>
>>>> Organization: bmp System Support - keep your systems running
>>>> Reply-To: bernard@bmpsystems.com
>>>> Date: Thu, 29 Mar 2001 17:25:30 +0200
>>>> To: netatalk-admins@umich.edu
>>>> Subject: Sherlock from Mac OS 9 slow
>>>> Resent-From: netatalk-admins@umich.edu
>>>> Resent-Date: Thu, 29 Mar 2001 10:34:45 -0500 (EST)
>>>>
>>>> Hi,
>>>>
>>>> I have a RedHat 6.2 Server running and while the server itself
>>>>is very fast,
> >>> Sherlock is extremly slow when doing searches from Mac OS 9 clients. The
> >>> server volumes are large, with about 50GB of data on every drive.
>>>>
>>>> I know there where problems with netatalk not supporting the FPCatSearch
>>>> function which Sherlock uses.
>>>>
>>>> Is this fixed in some version of netatalk out there on the net or does
>>>> someone know another solution to this problem?
>>>>
>>>> Bernard
>>>>
>>>> --
>>>> www.bmpsystems.com
>>>> +43 664 336 9196
This archive was generated by hypermail 2b28 : Sun Oct 14 2001 - 03:04:36 EDT