T3B Configuration options
Hello all,
I've inherited a deployed T3B running 3.1.5 firmware, and I'm trying to determine whether it's optimally configured for the application (a database-driven calendaring system).
<div class="pre"><pre>sunt3b:/:<33>vol listvolumecapacity raid datastandbyv0 134.890 GB1 u1d01-08u1d09sunt3b:/:<34>vol mode v0volume mounted cachemirrorv0 yeswritebehind offsunt3b:/:<35>sys listcontroller : 2.0blocksize : 64kcache : automirror : automp_support : nonenaca: offrd_ahead: offrecon_rate : medsys memsize: 128 MBytescache memsize: 1024 MBytesenable_volslice: offfc_topology: autofc_speed: 1Gbdisk_scrubber: onondg: befit</pre></div>
The application vendor recommends "stripe-sized blocks" of 8k, but I see that the T3B is configured for blocksize=64k. Is blocksize the same thing? (I see the T3's manual has discussion of a "cache block" which is 1/8 the size of blocksize, so blocksize=64k makes a cache block of 8k.)
Secondly, what is the precise behavior of the rd_ahead setting? The documentation isn't clear:
<table border="0" align="center" width="90%" cellpadding="3" cellspacing="1"><tr><td class="SmallText"><b>Quote:</b></td></tr><tr>& lt;td class="quote">
Setting read caching can reduce disk I/O by reading data for future retrieval. This refers to the number of sequential reads in a chunk of data. The two cache readahead threshold options are on and off. The default setting for the array is on.
</td></tr></table><table border="0" align="center" width="90%" cellpadding="3" cellspacing="1"><tr><td class="SmallText"><b>Quote:</b></td></tr><tr>& lt;td class="quote">
Sets the number of sequential read commands before automatically
switching to datablock read ahead. Set to off to always perform
datablock read ahead.
</td></tr></table>
Thank you!
[2052 byte] By [
atlauren] at [2007-11-25 23:03:06]
# 1
Looking at the config that you have vol0 is pretty much the entire array and you have it configured as raid 1(mirror) over disks 1-8 and disk 9 as a standby.
Now personally the T3 is meant as a raid 5 array as it has the onboard processing for the raid 5 calculations. I think you would see a big performance benefit if you changed the config of the array to raid 5 over disks 1-8 and leave disk 9 as a standby.
It is good to see that you have version 3 firmware on this as well.
# 3
<table border="0" align="center" width="90%" cellpadding="3" cellspacing="1"><tr><td class="SmallText"><b>m-lennon wrote on Wed, 22 February 2006 03:11</b></td></tr><tr><td class="quote">
I agree with Stuart, you will double your storage volume by changing to RAID 5. If your calander application is just mainly reads, RAID 5 should perform well. Search the forum for RAID 5 performance on T3, there have been interesting posts on this before.
</td></tr></table>
It's an Oracle Calendar installation. I should note, however, that it does *not* use Oracle's core database. It's an acquired product, and still uses the embedded database engine that came with the acquisition. The application's documentation explicitly recommends 1+0 configurations. The data storage needs are actually quite small, so I'm mainly looking for performance in this case.
Any ideas on the blocksize and rd_ahead settings?
# 5
You can stripe a number of T3 RAID 1 LUNs with SVM or QFS. We are currently experimenting with this setup,
but I haven't got any realistic performance statistics yet.
The only thing I could find that was worth reading was a blueprint:
<a href="http://www.sun.com/blueprints/1001/t3bp.pdf" target="_blank">http://www.sun.com/blueprints/1001/t3bp.pdf</a>
On page 10 the author explains what the rd_ahead parameter is for.
I would recommend going through the entire document as well.
