Re: vinum and superblocks.
It seems Jaye Mathisen wrote: Not sure what the right thing to do here is, or even if it's a real problem, but: I have 8 75GB IBM drives striped in a big raid 0 for monkeying with. newfs -i 131072 -v /dev/vinum/bighonkindisk seems to very nicely put all the data that newfs write out on to the first disk... It least, only the first disk gets any io, accoring to systat and iostat. Which would seem to me to be problematic in terms of using fsck -b, and also just for the fact that it would seem that you would have to hit that disk more often than the others, even though it's striped. I realize there's no protection with the raid 0, but the load doesn't seem evenly distributed on a transaction basis, even though the data is evenly spread. What's the right thing to do here? You need to adjust your stripe size so that the superblocks etc are not all put on the same disk, ie some odd stripe size is often best to distribute those to hopefully all disks... -Søren To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: vinum and superblocks.
On Fri, 30 Jun 2000, Soren Schmidt wrote: It seems Jaye Mathisen wrote: I have 8 75GB IBM drives striped in a big raid 0 for monkeying with. newfs -i 131072 -v /dev/vinum/bighonkindisk seems to very nicely put all the data that newfs write out on to the first disk... It least, only the first disk gets any io, accoring to systat and iostat. You need to adjust your stripe size so that the superblocks etc are not all put on the same disk, ie some odd stripe size is often best to distribute those to hopefully all disks... By 'odd', you mean small, Søren? If this is the case, be careful not to make the size too small, which will degrade the perfomance because of the Jaye, are you sure you doing a large (preferrebly at least one GB) sequential read or write from/to your disks? If you are only doing small random I/O, this could result in only one disk getting hit. But it does sound strange that when making the filsystem, only the first disk gets initialized.. Are you positively sure about this? The size of the vinum volume corresponds to approx 8 * 75GB ? Best regards Rasmus To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: vinum and superblocks.
Rasmus Skaarup wrote: By 'odd', you mean small, Søren? If this is the case, be careful not to make the size too small, which will degrade the perfomance because of the .. beacuse of the drive geometry. Best regards Rasmus To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: vinum and superblocks.
On Friday, 30 June 2000 at 9:16:33 +0200, Rasmus Skaarup wrote: On Fri, 30 Jun 2000, Soren Schmidt wrote: It seems Jaye Mathisen wrote: I have 8 75GB IBM drives striped in a big raid 0 for monkeying with. newfs -i 131072 -v /dev/vinum/bighonkindisk seems to very nicely put all the data that newfs write out on to the first disk... It least, only the first disk gets any io, accoring to systat and iostat. You need to adjust your stripe size so that the superblocks etc are not all put on the same disk, ie some odd stripe size is often best to distribute those to hopefully all disks... By 'odd', you mean small, Søren? If this is the case, be careful not to make the size too small, which will degrade the perfomance because of the Of the? But yes, you don't want stripes of less than about 256 kB. Jaye, are you sure you doing a large (preferrebly at least one GB) sequential read or write from/to your disks? If you are only doing small random I/O, this could result in only one disk getting hit. But it does sound strange that when making the filsystem, only the first disk gets initialized.. Are you positively sure about this? It sounds right to me. It seems I can't find the text I had intended to put in the man page; I'll follow up with this some time soon. But indeed, the access pattern for newfs will write only to the first subdisk if the stripe size is, say, 512 kB. I'm currently recommending odd stripe sizes like 273 kB; calculate something which will put superblocks 32 MB apart onto consecutive subdisks. Greg -- Finger [EMAIL PROTECTED] for PGP public key See complete headers for address and phone numbers To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message
Re: vinum and superblocks.
On Fri, Jun 30, 2000 at 09:16:33AM +0200, Rasmus Skaarup wrote: By 'odd', you mean small, Søren? If this is the case, be careful not to make the size too small, which will degrade the perfomance because of the I presumed he ment 'odd' as in 'not even' David. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-hackers" in the body of the message