From mboxrd@z Thu Jan 1 00:00:00 1970 Message-ID: Date: Mon, 28 Jul 2008 17:31:22 -0400 From: "Russ Cox" To: pavelcak@gmail.com, "Fans of the OS Plan 9 from Bell Labs" <9fans@9fans.net> Subject: Re: [9fans] Venti arenas and index guidelines In-Reply-To: <6256dcf20807281418h55ffe4d8o1eadae8d807df3bd@mail.gmail.com> MIME-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Content-Disposition: inline References: <6256dcf20807281418h55ffe4d8o1eadae8d807df3bd@mail.gmail.com> Topicbox-Message-UUID: f3a0ad8c-ead3-11e9-9d60-3106f5b1d025 > I have a fossil on an 18G hard drive, which from what I've seen, is > probably a much larger fossil than I need. However, I wasn't planning > on changing it unless it's really trivially easy to do so. I also have > another 18G drive and two 50G drives. I thought the thing to do would > be to mirror the 50G drives for venti arenas and that would leave the > 18G drive for isects (and bloom?). Venti is much less sensitive to disk seek times than it once was. (See http://swtch.com/~rsc/papers/fndn/ and s/Foundation/Venti/g.) So to a first approximation, it basically doesn't matter what you do. Mirroring is a good idea; that's a good argument for putting the data on the 50G pair. You could also put the isect and bloom filter there, to keep the setup simple. You only need 5% = 2.5G for index; 18G is overkill. The bloom filter helps most when you are first initializing the venti, and for that it can be small (64MB is probably plenty). Russ