Re: How to read files from disk directly?
Re: How to read files from disk directly?
- Subject: Re: How to read files from disk directly?
- From: Wim Lewis <email@hidden>
- Date: Tue, 05 Jul 2011 16:53:21 -0700
On 4 Jul 2011, at 8:55 AM, Gregg Wonderly wrote:
> One of the things that I'd be tempted to investigate is disk head scheduling under high load. Does HFS utilize a scheduler that might help mitigate some of the "seek" issues if you have enough outstanding requests that the head motion and read buffering can be optimized in the kernel, for you?
>
> If so, try a work queue with 1, 5, 10, 25, 50, 100, 200 threads and see if a simple opendir/readdir traversal with work items fed to such a queue might provide a reasonable kernel based optimization stream.
This was going to be my suggestion as well.
Also, I don't remember whether HFS+ works this way, but on some filesystems the inode number corresponds to its location on disk, and file contents are preferentially stored near their inodes; if you sort your work queue by inode, that might increase locality of reads from the disk.
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Filesystem-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden