Bruce,
I called it "folklore" because I was told about the supposed improvement in disk scrub speed by a couple of Netapp engineers at a recent engineering meeting I attended. Maybe it would have been better to say "from unofficial sources, in a casual conversation, sharing afternoon drinks at a Netapp event".
The reconstruct speed is set at the default of 4. The 2 systems are very lightly loaded from a user traffic standpoint, cpu load at under 10% during reads and bursting to 20% during WAFL writes. There are 5 minute periods every 30 minutes when the cpu goes up to 95% during SnapMirror operations. Until there is considerably more user traffic on the F840's, I will not be able to ascertain user impact. My main comment was about the total time to do the scrub. Of course, time isn't the key factor - impact on user file activity would be much more relevant, and I don't have numbers for that yet.
-----Original Message----- From: Bruce Sterling Woodcock [mailto:sirbruce@ix.netcom.com] Sent: Monday, September 25, 2000 6:36 PM To: Sam Schorr; 'Todd C. Merrill' Cc: toasters@mathworks.com Subject: Re: F840
One piece of reality-versus-engineering-claims relates to disk scrubbing. We turned
off
disk scrub on the F760's because it really impacted user file access. I
was
told this issue would no longer be a problem. Well, it takes the F840's, each with 12 shelves of 36 gig drives, from 1AM to 2PM (yes, 2 PM) to
scrub
under very light load.
And the impact on performance was?
And your scrub speed was set to?
Another piece of folklore I guess.
What folklore? It's not clear from the text that the engineering claim was wrong.
Bruce
Bruce,
I called it "folklore" because I was told about the supposed improvement
in
disk scrub speed by a couple of Netapp engineers at a recent engineering meeting I attended. Maybe it would have been better to say "from
unofficial
sources, in a casual conversation, sharing afternoon drinks at a Netapp event".
But this wasn't what you said. You said:
"We turned off disk scrub on the F760's because it really impacted user file access. I was told this issue would no longer be a problem."
This doesn't seem to be a claim about the speed of disk scrub. It seems to be a claim about the impact on user file access.
The reconstruct speed is set at the default of 4. The 2 systems are very lightly loaded from a user traffic standpoint, cpu load at under 10%
during
reads and bursting to 20% during WAFL writes. There are 5 minute periods every 30 minutes when the cpu goes up to 95% during SnapMirror operations. Until there is considerably more user traffic on the F840's, I will not be able to ascertain user impact. My main comment was about the total time
to
do the scrub. Of course, time isn't the key factor - impact on user file activity would be much more relevant, and I don't have numbers for that
yet.
Okay. Personally, I don't think 13 hours of disk scrubbing, once a week, is bad at all, particularly if it does not impact user file activity noticeably. But what I was getting at was you seemed to make a claim about the file access first, then provided evidence for how long it took.
Bruce
-
Bruce Sterling Woodcock -
Sam Schorr