[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CA+55aFybtRdg=AzcHv3CPm-_wx8LT2_CXaKr4K+i94QSPauZOw@mail.gmail.com>
Date: Thu, 2 Aug 2012 13:32:41 -0700
From: Linus Torvalds <torvalds@...ux-foundation.org>
To: Josh Triplett <josh@...htriplett.org>
Cc: "Eric W. Biederman" <ebiederm@...ssion.com>,
Sasha Levin <levinsasha928@...il.com>,
Tejun Heo <tj@...nel.org>, akpm@...ux-foundation.org,
linux-kernel@...r.kernel.org, linux-mm@...ck.org,
paul.gortmaker@...driver.com
Subject: Re: [RFC 1/4] hashtable: introduce a small and naive hashtable
On Thu, Aug 2, 2012 at 1:25 PM, Josh Triplett <josh@...htriplett.org> wrote:
>
> Sorry, I should clarify what I meant: you'll have a total of one extra
> indirection, not two.
Yes. But the hash table address generation is noticeably bigger and
slower due to the non-fixed size too.
In general, you can basically think of a dynamic hash table as always
having one extra entry in the hash chains. Sure, the base address
*may* cache well, but on the other hand, a smaller static hash table
caches better than a big one, so you lose some and you win some.
According to my numbers, you win a lot more than you lose.
> Does your two-level dcache handle eviction?
>
> Mind posting the WIP patches?
Attached. It's against an older kernel, but I suspect it still applies
cleanly. The patch is certainly simple, but note the warning (you can
*run* it, though - the race is almost entirely theoretical, so you can
get numbers without ever seeing it)
Linus
Download attachment "patch.diff" of type "application/octet-stream" (6855 bytes)
Powered by blists - more mailing lists