From mboxrd@z Thu Jan 1 00:00:00 1970 From: Sander Subject: Re: New idea about RAID and SSD Date: Tue, 1 Sep 2009 15:05:46 +0200 Message-ID: <20090901130546.GA10305@cumulus> References: <4A9D1746.1020407@mmmm.it> Reply-To: sander@humilis.net Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: linux-btrfs@vger.kernel.org To: Massimo Maggi Return-path: In-Reply-To: <4A9D1746.1020407@mmmm.it> List-ID: Hello Massimo, Massimo Maggi wrote (ao): > SSDs have low latency but a high price per GB, > Traditional hard disks have high latency, but high sequential read/write > speed and low price per GB. > Is possibile to use a SSD for metadata, which requires many seeks and is > relatively small, in a special "RAID mode" with a traditional hard disk > for the extents of the real data? > A cheap but performant SSD (maybe 32 GB) + a big and fast HD (maybe 1.5 > TB, or two in RAID0 - 3TB ), wouldn't create an array much cheaper than > a ssd-only array of the same size, and much faster (in > not-only-sequential workload) than one or two traditional HDs in RAID0? > Would it work? If you talk RAID0 (eg no redundancy), you could RAID0 one or several traditional disks, and use the SSD as a journal device. That would be ext3/4 only btw. With mdadm you could create a RAID1 and use --write-mostly: -W, --write-mostly subsequent devices listed in a --build, --create, or --add com- mand will be flagged as 'write-mostly'. This is valid for RAID1 only and means that the 'md' driver will avoid reading from these devices if at all possible. This can be useful if mirror- ing over a slow link. Where the 'slow link' would be the traditional disk. But this is raid1 and doesn't help in your case (but couldn't resist the need to mention it :-) Sander -- Humilis IT Services and Solutions http://www.humilis.net