atom feed16 messages in org.freebsd.freebsd-stableRe: Dump takes ages on SCSI. (severa...
FromSent OnAttachments
David GilbertJul 2, 2000 1:24 pm 
Kent StewartJul 2, 2000 2:53 pm 
David GilbertJul 2, 2000 5:42 pm 
Kent StewartJul 2, 2000 6:50 pm 
Chad R. LarsonJul 2, 2000 8:44 pm 
David GilbertJul 2, 2000 9:25 pm 
Kenneth D. MerryJul 2, 2000 10:55 pm 
David MaloneJul 3, 2000 12:01 am 
kostikJul 3, 2000 1:35 am 
Gerhard SittigJul 3, 2000 10:38 am 
Mike SmithJul 3, 2000 6:57 pm 
Mike SmithJul 4, 2000 4:08 am 
Mike SmithJul 4, 2000 5:22 am 
Gerhard SittigJul 4, 2000 10:16 am 
Mike SmithJul 6, 2000 2:36 am 
Don LewisJul 14, 2000 1:36 am 
Subject:Re: Dump takes ages on SCSI. (several times faster on IDE)
From:David Malone (dwma@maths.tcd.ie)
Date:Jul 3, 2000 12:01:31 am
List:org.freebsd.freebsd-stable

On Sun, Jul 02, 2000 at 11:55:24PM -0600, Kenneth D. Merry wrote:

When the da(4) driver does a dump, it uses the CAM transport layer's polled completion/action facility, xpt_polled_action().

It looks like there is a substantial delay (1ms) in that routine between poll intervals.

If you'd like it to go faster, you'll most likely have to ask Justin about it. (i.e., I didn't write it.)

I asked Justin about this about a year ago. We have a machine with 1GB of ram, which was dieing with NFS panics and taking about 20/30 minutes to dump. It's nothing to do with the delay (I tried reducing it and it went at exactly the same speed). Some drives just don't like been driven in whatever polled mode is used - I don't actually think I've seen any SCSI drives that dump quickly (including various Quantum and Seagate disks) and , but Justin said that he had some which did.

I thought about changing the da dump code to write in 64k or bigger chunks to avoid the problem, but never got around to it.

David.

To Unsubscribe: send mail to majo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message