UNSOLVED
monitoring forced flushes
I'm kind of new at this. I've got a CX 700. What is the best way to monitor or find out what the forced flush rate is for the whole SAN? I see I can choose each LUN and see the number of forced flushes. Is that for the whole SAN or do I need to add up each LUN's Forced Flushes to find out the total number of forced flushes on the SAN
Thanks
Thanks
485
0
Posted May 10th, 2009 07:00
"navicli -h x.x.x.x getcache" show cache details; But does not look real time486
0
Posted May 11th, 2009 03:00
You could consider using Navisphere Analyzer ! Lots of details in thereIf you look up detaisl for a specific LUN, it's for that LUN only. You'll need to add up all details for all LUN's to get a total. If you use Analyzer, you'll get much more ways to get stats.
But.... out of curiousity: why do you call your Clariion a SAN ? The SAN is the network the Clariion is conencted to, so the FC switches ! The Clariion is called Storage Array and it's connected to the SAN with a few fiber patch cables.
486
0
Posted May 11th, 2009 06:00
Thanks, I am using Navisphere Analyzer to get the Forced Flushes/S rate. So I guess that if I see 50 FF/sec on one Lun and another 40 FF/Sec on another LUN, then, assuming these are the only 2 LUNs, that would equal 90 FF/sec overall on the whole storage array (Thanks for clarifying between Storage array and SAN - makes sense!).486
0
Posted May 11th, 2009 07:00
A better way to look at force flushing for the whole array is %Dirty Pages. Force Flushing begins when the Write Cache% Dirty Pages exceeds the High Watermark. If your watermarks are set to 60/80 (the normal default), when % Dirty Pages exceeds 80%, the cache will start to force flush - typically on the LUN(s) causing the cache to get filled up, but also other LUNs that are close to 80% full.
If %Dirty Pages gets to 99% full, then the cache is completely full and can accept no more Writes - then the whole write cache will start to force flush - you want to avoid this if possible as it will slow down everyone using the array.
Check the On-Line Help - there is a section specifically for Analyzer that has a lot of good information.
glen486
0
Posted May 11th, 2009 10:00
this is exactly what I was looking for. Thanks for your help. I actually have a PDF document on Navi Analyzer, but I'll check on-line help for more info.486
0
Posted May 12th, 2009 01:00
Hey Steven,
I'm glad you got your answer and I noticed you marked your question as "answered". If there's a post that helped you in particular or partial, could you please mark these post as such ?486
0
Posted May 12th, 2009 06:00
Quck and dirty way:
Right click on the array go to properties. Click on the Cache tab. At the bottom you will see Percent dirty pages. if you see this hitting 99 percent you are getting forced flushing. All the other ways are definately better than this because you get the historical data
Hope this helps.
Mike486
0
Posted February 12th, 2010 17:00
All,
I was looking at some navi logs over a 5 day period and noticed having consistent high watermarks...mostly in the upper 70s. There were a few 90s and one was at 99. My watermark is currently using the default settings of 80-60 and I was wondering if I should change it to 60-40 to see if I can get some performance on my hosts.
If I understood correctly, if the dirty pages hit 99%, that write cache is disabled and that the SP will perform a 'forced' flush until it hits the low watermark. And i know if it hits above 80 that it will also performs a forced flush. Let's say it was at 81%...does that mean that cache is also disabled until it hits 60 or does that only pertain to when dirty pages hits 99% because there's no room?
At what point, after looking at the logs, dictates whether I should drop my watermark settings? For example, if I'm seeing consistent dirty pages at the high 70% levels, and if I'm considering dropping the watermark to 60-40, wouldn't I be seeing more forced flushes? And isn't several forced flushes bad?
81
0
Posted February 13th, 2010 03:00
I wouldn't lower the low and high watermarks. If the HW is set to 60 and cache usage is over 60%, forced flushing will begin sooner and this might have a more serious performance impact. I'd suggest to experiment to set the LW and HW to 90/70. This way the cache is used more effectively. You will see more forced flushes though, but that flushing is over more quickly as well, since the LW is higher as well
I have seen people experimenting with values of 50/70, but that was in specific database environments. I know there are threads on the forums about this.
Suggestions anyone ?
skong
14 Posts
485
0
Posted May 10th, 2009 04:00
Thanks again!