cancel
Showing results for 
Search instead for 
Did you mean: 

Kernel parameters

David Valentino_1
Occasional Advisor

Kernel parameters

We are running Tru64 unix 5.1b. We are running Oracle/SCT/Banner applications. We migrated from an Alpha ES40 to a cluster with two Alpha 4100s. We forgot about the kernel parameters on the cluster. We went from Banner 5 to Banner 6. Users are experiencing slowness in the applictaion. Should I change these parameters to match the ES40? I have included a file that shows the differences. Thank you.
7 REPLIES
Johan Brusche
Honored Contributor

Re: Kernel parameters


There's a mismatch on the AS4100 between per_proc_data_size and shm_max, .._data_size should be lager than shm_max, eg 2200000000

You do not mention "new_wire_method" in "vm:" and "fifo_do_adaptive" in "vfs:", but you better set both to =0 instead of the default of 1.

Johan.

_JB_
Michael Schulte zur Sur
Honored Contributor

Re: Kernel parameters

Hi David,

I do not know the software, you are referring to. But the values don't look so bad, if you are not running into resource errors. As much as I know, a ES40 is faster than a single 4100. You would have to share more about the software/hardware configuration in the cluster to see, if anything hinders the machines.

Michael
Joris Denayer
Respected Contributor

Re: Kernel parameters

David,

You can also find more info about Tru64UNIX tuning for Oracle in the following document.

http://h30097.www3.hp.com/docs/base_doc/DOCUMENTATION/V51B_HTML/ARH9GCTE/TITLE.HTM

Joris
To err is human, but to really faul things up requires a computer
Hein van den Heuvel
Honored Contributor

Re: Kernel parameters


An Alpha 4100 is in general slower than and ES40. What are the CPU speeds on the systems?
What are the chips? Are they both at least Ev5.6... because recent oracle versions will use the Byte-Word instructions which were not available on Ev5 (300 Mhz) chip (someone correct me on the details, I don't have them handy!).


Instead of looking at system parameters I would much rather see you focus on performance observations and Oracle params.

Can you quantify 'slowness'. Did you run 'collect' or at least a simple iostat or vmsstat to get an impression on IO/Mem/Cpu usage for the system before and after? What is the current average cpu use with the slow performance?

The memory on the new systems us also lower.
Is the SGA still the same size? Does you have enought memory to hpld SGA + processes?

hth,
Hein.
Ralf Puchner
Honored Contributor

Re: Kernel parameters

each machine must be tuned regarding the available resources. Have a look into the oracle or application tuning guide for calculating the correct values.
Help() { FirstReadManual(urgently); Go_to_it;; }

Re: Kernel parameters

This is my way to calculate some critical values per node:

First I use really big values,
msg_mni=512
sem_mni=512
shm_min=512
shm_max= About half of the physical memory max_process_per_user=2048 max_threads_per_user=2048
ubc_minpercent = 3
ubc_borrowpercent = 3
ubc_maxpercent = 60


When all services are up and running and used then you can count current values:

Minimum values are
MSG_MNI
ipcs -q|wc -l

SEM_MNI
ipcs -s|wc -l

SHM_MNI
ipcs -m|wc -l


MAX_PROCESS_PER_USER
ps -ef|wc -l

All the running process ~ max_process_per_user

MAX_THREADS_PER_USER
expr `ps -efm|wc -l` - `ps -ef|wc -l`

All the running threads ~ max_threads_per_user

Remember:
The root-user does'n have limit on the process and threads!


If you use very big values, start to thing that everything is in the big tables witch fragmented ie. too big values means overhead, less (in memory) users and so on.

The values what you start the use are more then the sum of every node current values.
-----------------------------------

SHM_MAX
About half of the physical memory or
Oracle9i : SHM_MAX=4G
oracle 8i : SHM_MAX=2G


UBC
ubc_minpercent = 3
ubc_borrowpercent = 3
ubc_maxpercent = 60

If you have a big physical memory then calculate how much memory is a 3%, this gives to you image how muct data stays in the memory to waiting flush!

Yes,
I know, all of you says, this values are to small.. but you must trust and test
David Valentino_1
Occasional Advisor

Re: Kernel parameters

Thank you for all of your replies. We just recently brought the ES40 into the cluster and are now using it as the master server. We'll adjust the parameters on the Alpha 4100.