Archives of the TeradataForum

Message Posted: Thu, 17 Nov 2005 @ 10:29:08 GMT

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

CPU:

The Ferret Master task is sitting in FGETS waiting for User input.

     Backtrace for process 1143 (/tpasw/bin/ferret)
     00: 0xfe0008f0 0xfe000940 resume
     01: 0xfe000948 0xfe000984 __AIR_schsleep+0x11e
     02: 0xfe00098c 0xfe000a40 __AIR_msgrxmain+0x1c3
     03: 0xfe000a48 0xfe000cf0 __AIR_msgrx+0x39
     04: 0xfe000cf8 0xfe000d5c spec_read+0x294
     05: 0xfe000d64 0xfe000dd0 vop_read+0x32
     06: 0xfe000dd8 0xfe000e24 systrap+0x116
     07:            0xfe000e34 sys_call+0x3f from 0xbffcd5f5

under the cover this FGETS actually turns into a MSGRX waiting for CNS to send it a message to satisfy the read.

All The Background tasks are sitting on a Mailbox waiting for the Master task to send them a message

     Backtrace for process 1145 (/tpasw/bin/ferret -s -g 0x00216944)
     00: 0xfe000bc0 0xfe000c10 resume
     01: 0xfe000c18 0xfe000c54 __AIR_schsleep+0x11e
     02: 0xfe000c5c 0xfe000d10 __AIR_msgrxmain+0x1c3
     03: 0xfe000d18 0xfe000d90 __AIR_msgrxseg+0x11b
     04: 0xfe000d98 0xfe000e20 __AIR_sysnscall+0x53a
     05:            0xfe000e34 hfcsyscall+0x42 from 0xbfc8cfed

Sleeping on a mailbox waits for a Synchronous event. there is no polling.

They can sit in this state for Days and not accumulate any CPU time. I started composing email this at 2:00 and started up a couple ferret tasks. I got side tracked and finally got back to this mailgram.

     {mybox} date ;  ps -ef | egrep ferret
     Wed Nov 16 17:52:48 PST 2005
         root  1614     1  0 14:13:49 ?        0:00 /tpasw/bin/ferret -s -g 0x00216944
         root  1143     1  0 14:13:49 ?        0:00 /tpasw/bin/ferret
         root  1145     1  0 14:13:49 ?        0:00 /tpasw/bin/ferret -s -g 0x00216944

3 1/2 hours later my Ferret tasks have used 0 seconds worth of CPU time.

Memory :

Everything Ferret has mapped in is Shared with other processes (The Gdos, the FSG segments, executable code) or Swappable/Pageable(sp). (its data area, The Scratch segments, The Stacks, the break/malloc area)

There maybe a little section of the PROC area that isn't swappable just so the system still knows the process still exists.

We have 1000 Awtmain and SqlDriver Processes Running on a typical TPA node. 10 idle Ferret tasks will quickly be paged out to make way for the important work.

Therefore Ferret uses no CPU and has the potential to use no Memory if the system decides the memory is needed for better purpose while Ferret is Idle.

Now if for example the ferret master task is getting woken up incorrectly because it thinks it got Input from the user because of a broken Pipe or something and is constantly looping trying to process, read, process, read, then yes it could use a lot of resources....

But then it is NOT idle. You might not be using ferret directly (a perfectly valid definition of idle) but Ferret is doing something in the background that it isn't supposed to be doing and we need to fix it.

If that is the case, open an incident with the GSC and lets see if we can get it fixed.

Now it is possible that the existence of the Ferret Process May cause Another Task maybe CNS for example to use additional resources managing the Window. I don't know that code as well as I know Ferret.

If there is a resource issue somewhere else just because the Ferret tasks exist, open an incident with the GSC and document what you are seeing so we can try to reproduce it in house and get these issues fixed.

The Bottom line is we can't fix something unless we know it is broke and how to verify that we fixed the problem.

And while we are the subject of Issues....Not all issues are problems. If you have an idea or suggestion on how to make Teradata better you can fill out the Partners Enhancement request form and it will be brought before a committee of other customers who will review your Idea.

www.teradata.com/teradata-partners/about/enhance.shtml

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