Date: Sat, 2 Jun 2001 10:54:34 -0700 (PDT) From: Matt Dillon <dillon@earth.backplane.com> To: Ian Dowse <iedowse@maths.tcd.ie> Cc: freebsd-hackers@FreeBSD.ORG Subject: Re: UFS large directory performance Message-ID: <200106021754.f52HsYR04024@earth.backplane.com> References: <200106021207.aa78407@salmon.maths.tcd.ie>
next in thread | previous in thread | raw e-mail | index | archive | help
Oh yah, one more thing. At risk of making the second level block
more complex you might want to make a real structure out of it rather
then a simple array. Something like this:
struct whateverYouNameTheSecondLevelBLock {
short use;
char filler[sizeof(void *) - sizeof(short)];
secondLevel_t array[128];
};
We could maintain a use count on the second level block which
would allow us to make better decisions on which one to reuse.
You could get more complex and keep the second level blocks on
their own LRU queue, making reclaims trivial. This would require
a doubly linked list node (lruq) and a backptr to allow you to
unlink the second level block from the first level array, e.g.
'*secondLevelBlock->backptr = (void *)-1;'
struct whateverYouNameTheSecondLevelBLock {
TAILQ_ENTRY(blah) lruq;
firstLevel_t *backptr;
short use;
char filler[sizeof(void *) - sizeof(short)];
secondLevel_t array[128];
};
For now I would recommend the first approach (you could add a queue node
to the summary structure to allow scanning all the summary structures).
This second approach would be a good second-round approach.
-Matt
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200106021754.f52HsYR04024>