Re: disks for redo logs

From: joel garry <>
Date: Sun, 15 Feb 2009 16:24:41 -0800 (PST)
Message-ID: <>

On Feb 13, 4:37 am, Noons <> wrote:
> joel garry wrote,on my timestamp of 13/02/2009 5:51 AM:
> > I've seen this advice before, but haven't done anything, since, hey,
> > if the thing is I/O bound, wouldn't making I/O initiates more often
> > make it worse?
> No, not more often. Jacking up the priority of lgwr does nothing to change the
> frequency or rate of lgwr IO activation: that is not determined by OS priority.
> What it does is this: *if* or *when* the lgwr is posted to wake up and start an
> IO, then it does so without having to sit in a queue with other processes at
> same priority who might also be waiting for CPU time slot.
> Once the IO is started, lgwr simply goes back to sleep waiting for IO completion
> or next activation, without waiting for the end of time slot.
> Because lgwr doesn't have to do much other than start a write of buffers to
> disk, it won't consume much CPU before it suspends waiting for completion.
> As such, the higher priority does not significantly affect other processes.

Well, I have two situations at different times: CPU bound and waits on lgwr. I realized after I posted I didn't really answer your question, because I got distracted by posting the middle-of-the-night waits. To answer your question, at random times I get up to about 2M writes waiting a couple seconds. I've just buried my head in the sand and mumbled something about hp-ux housekeeping, since I don't have permission to really analyze what is going on. I'm hoping to sneak in some glancing when I do a standby configuration soon.

> Classical OS scheduling theory: IO-bound processes get a higher priority so they
> can quickly start their short CPU demand to initiate an IO, with the least
> possible delay, and then go back to sleep.
> > documented limits. Needless to say, the latter has gotten much worse
> > with scaling and going to fewer, faster Itanium processors.
> Bingo! It never fails. Unreal...
> > so ask for Oracle trace files. Which of course don't say anything at
> > all because the program isn't asking the database for anything when it
> > is cpu bound. I wouldn't want to make that worse fighting background
> > processes.
> Ah, OK: not much redo log generation by all that in-memory crap. Then you're OK.
> Here, I'm seeing redos sized at 200MB switching every 5 seconds on a good day,
> for periods in excess of one hour. And they gotta be archived...

Why so small?

> > Unfortunately, some end-of-month processing has been converted to use
> > those misfeatures in the new version, so I get a lot of
> > "arewethereyetarewethereyetarewethereyettheoldsystemwasfaster!" Users
> > having forgotten not so long ago when it took days so they called me
> > in... 8 years already? Wore out a whole Chrysler...
> Yikes! Man! How many miles are you doing daily?
> When I was with the air force, at one stage I was doing 55 each way...

Well, by buying an "Irvine car" (New Beetle driven by a little old lady on Sundays to be a docent at a local adobe - ripping up the undersides on the dirt roads in the process, and after a year I haven't gotten her dog's smell out) I take the train most of the time these days. But it was 55 miles each way, until we switched sites and it became 62.5 miles each way. Funnily enough, it was also 55 miles each way when I worked for mil, but in the other direction, much more horrid traffic. This current commute is only bad the last 10 miles or so, I don't mind so much cruising at 80 next to the ocean. But it's often nicer on the train cruising at 80 next to the ocean, reading the newspaper and pretending not to ogle the college girls. Either way gets messed up every few months - someone offs themselves on the tracks, train engines blow oil lines, gasoline truck flips on the freeway next to the nuke plant, etc. Friday nights are standing room only either method, I take an earlier train that's not so bad. is an interesting traffic site. I actually considered getting a Bentley (much like DKB got an RR, I only noticed that posting today), but when it comes down to it, I still wind up pounding the crap out of cars on the freeways, even taking the train most of the time. Wife's car with 5K miles on it got munched on, yes, Friday the 13th, around the corner from her university teaching job. Hit by uninsured lady from Persia.

The lifestyle and pay scales are different enough between where I live and work to make the commute worth it. I can work from home sometimes, too. I've managed to get things to the point of not having to work nights or go in on weekends hardly ever. Prolly just jinxed it :-)


-- is bogus.
Received on Sun Feb 15 2009 - 18:24:41 CST

Original text of this message