Archives of the TeradataForum

Message Posted: Tue, 20 Dec 2005 @ 10:57:34 GMT

	<Prev	Next>		<<First	<Prev	Next>	Last>>

There are 2 Values in the DBS control GDO which affect Minicylpacks. I made sure these values existed in V2r3.

     dbscontrol

     display  filesys

         2.  MiniCylPackLowCylProd    =  ( default 10 )
         8.  Cylinders Saved for PERM =  ( default 10 )

you can change these value to something higher if you want to be more aggressive.

Minicylpacks occur when an AMP has less than

(MiniCylPackLowCylProd) cylinders FREE if it is a PERM request. ( insert into a BASE table )

(MiniCylPackLowCylProd + Cylinders Saved for PERM) FREE cylinder if it is SPOOL request. ( a Select type query )

Also anytime you allocate cylinders , PERM or SPOOL , could cause a MiniCylpack if the number of FREE Cylinders falls to less than the threshold value of FREE cylinder.

Therefore Loads, Selects, Collect stats.... Anything that needs to allocate a cylinder can cause a MiniCylpack if the number of FREE Cylinders falls to less than the threshold value of FREE cylinder.

MiniCylpack runs individually on each amp and therefore SKEW could cause one amp to have less than the threshold value of FREE cylinder before another AMP.

I don't know of a way to identify the QUERY which is causing the MiniCylpack since it is timing related. We try to report the Host Id and Session number of the query which allocates the Threshold Cylinder if we can. Look for this in the MiniCylpack error message.

     For Host %d Session %d

However maybe it isn't that queries fault. for example...

Let say Query 1 uses all but 11 cylinders, and then Query 2 used 2. Query 2 will cause the MiniCylpack because it allocated the cylinder which caused us to fall below the threshold value of FREE cylinders. It wasn't really Query 2's fault. it was the fact Query 1 was using all the space.

In addition, We only report the Minicylpacks message every 5 minutes to avoid flooding the Event log. this is to avoid thrashing when we are just at the threshold value.

Therefore if the query is still running 5 minutes from when the Cylpack occurred, you might be able to see it with PMON or QRYSESSN, but again just because a Session allocated that Threshold cylinder and caused the MiniCylpack, doesn't mean it was that queries fault the AMP was full.

You can run SHOWSPACE from FERRET via Teradata manager and check the number of FREE cylinders you have on each AMP. Again if the number of Free Cylinders is approaching 20 or Less on an amp, it is a candidate for MiniCylpack to occur.

You might want to consider running PACKDISK on a regular basis to try and coalesce freespace within Cylinders allocated to PERM tables, back into Cylinder Sized chunks to allow your number of FREE cylinders to go up so you maintain a good pool of FREE cylinders for your Queries to use for SPOOL.

You should work with your NCR Site team to proactively manage the Disk space on your system to avoid Minicylpacks.

	<Prev	Next>		<<First	<Prev	Next>	Last>>