Skip site navigation (1)Skip section navigation (2)
Date:      Tue, 29 Apr 1997 15:32:04 +0200
From:      Poul-Henning Kamp <phk@dk.tfs.com>
To:        Michael Hancock <michaelh@cet.co.jp>
Cc:        Gary Palmer <gpalmer@freefall.FreeBSD.ORG>, David Greenman <dg@root.com>, Don Lewis <Don.Lewis@tsc.tdk.com>, fs@FreeBSD.org
Subject:   Re: the namei cache... 
Message-ID:  <2112.862320724@critter>
In-Reply-To: Your message of "Tue, 29 Apr 1997 21:09:47 %2B0900." <Pine.SV4.3.95.970429205108.7361C-100000@parkplace.cet.co.jp> 

next in thread | previous in thread | raw e-mail | index | archive | help
In message <Pine.SV4.3.95.970429205108.7361C-100000@parkplace.cet.co.jp>, Micha
el Hancock writes:
>On Mon, 28 Apr 1997, Gary Palmer wrote:
>
>> > However, addressing the performance needs of news servers or large
>> > /var/mail directories might lead to a lot of complications.  It's like you
>> > need some kind of dynamic hash table creation capability for each
>> > directory vnode.  I don't think this is practical. 
>> 
>> I don't think any form of non-hashed search will work well with this:
>> 
>> root@news:/var/spool/news/misc/jobs/offered> ls | wc -l
>>    20040

Well, the question is how many of them are in the namecache at one time.

Few as far as I can tell.

It may actually make sense for news servers to only cache directories
in the name cache...

>Poul's analysis regarding stale entries seems to make sense though, there
>would be an increase in the hit rate if we could efficiently move stale
>entries to the head of the LRU list or lists.  It'd be nice to be able to
>keep the hash around while solving this problem too though.

There's only one problem, it reduces the max length of names in the cache
by another 8 bytes :-(

>I'm going to try the Poul's patches in a week or so.  I just doing
>WorldStone on RELENG_2_2, it's not incredibly scientific but it's a
>nightly cron job anyway and RELENG_2_2 isn't changing very much so you can
>get somewhat of a controlled environment. 

WorldStone is not very good for this, or rather, "make depend" is just
as good.

I'm still playing with various strategies.  If anybody are willing to 
test and measure (with whatever apps you want) it will be much 
appreciated.

--
Poul-Henning Kamp           | phk@FreeBSD.ORG       FreeBSD Core-team.
http://www.freebsd.org/~phk | phk@login.dknet.dk    Private mailbox.
whois: [PHK]                | phk@tfs.com           TRW Financial Systems, Inc.
Power and ignorance is a disgusting cocktail.



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?2112.862320724>