Tru64 Unix 中的主題 sysconfigtab 之 max_sessions???

sysconfigtab 之 max_sessions???

watermelonyu — Tue, 07 Dec 2004 04:28:33 GMT

請問 Tru64 的 max_sessions 所設的值是做什麼用的?

與 Oracle init.ora 的 prcess 設定有何關連?

Thanks

sysconfigtab 之 max_sessions???

watermelonyu — Tue, 07 Dec 2004 07:15:17 GMT

Hello Eric:

詳細的資料還需要進一步尋找, 不過有一些guidelines

描述如下:

max_sessions

>>> At least the number of Oracle processes

plus 2.

如果有進一步資料, 將儘速提供.

Best Regards

Danny Lin

sysconfigtab 之 max_sessions???

watermelonyu — Tue, 07 Dec 2004 23:50:46 GMT

Hello Eric:

After search Oracle metalink, there is one doc

as follow:

http://metalink.oracle.com/metalink/plsql/ml2_documents.showDocument?p_database_id=NOT&p_id=181489.1

that has some information and tuning suggestion for max_sessions:

- max_sessions - Should be set to at least <# of Oracle processes + 20>, it should be <+20>

not <+2>

Best Regards

Danny

sysconfigtab 之 max_sessions???

watermelonyu — Wed, 08 Dec 2004 03:15:58 GMT

Hello Eric:

The following is reply from ITRC worldwide

forum:

Danny,

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

Go for the +20.

This is not a major (memory) cost.

It is just a 'fools guards'.

It is there such that a process/application can not go wild and reserve unlimited resources putting the rest of teh system at risk.

Just give a ballpark. One thousand more or less makes little or no difference. 2 or 20 more or less makes absolutely no difference.

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

Best Regards

Danny Lin

sysconfigtab 之 max_sessions???

watermelonyu — Wed, 08 Dec 2004 03:58:03 GMT

Hi Danny,

Thanks. 還是依照 Oracle 的建議加 20 好了.

But...can you explan the relation between RDG & Oracle INTER-INSTANCE performance?

Thanks so much

Eric

sysconfigtab 之 max_sessions???

watermelonyu — Wed, 08 Dec 2004 11:53:16 GMT

Hello Eric:

More information for you:

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

Ya sure there is a relation between rdg and oracle nod-to-node communication. But max_sessions is not going to make a speed difference. RDG is designed to exploit the memory channel to provide a low latency (microseconds) connection between systems. The bandwidth is decend, the latency great. This is critical for the lock management portion of the internode communication. Many small packages going back and forward quickly. RDG is best there, gigabit ethernet not too bad.

The other component of internode traffic is db block shipping for the cache coherency (cache fusion). For that traffic bandwith is more important than latency and gigabit ethernet may outperform the memory channel.

Try to get hold of the rdgstat utility?! or just use sysconfig -q rdg to get some stats?

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

This should be related to Oracle 9iRAC or OPS

of Oracle 8.

Best Regards

Danny Lin

sysconfigtab 之 max_sessions???

watermelonyu — Thu, 09 Dec 2004 00:07:08 GMT

Hello:

As I mentioned that this parameter has some

relationship with INTER-INSTANCE PERFORMANCE for

your Oracle 9iRAC or Oracle 8 OPS, so I post the

follwing attachment for your reference, if you

still have performace, let your DBA call Oracle

for assistance.

sysconfigtab 之 max_sessions???

watermelonyu — Thu, 09 Dec 2004 01:09:49 GMT

Dear Danny,

As you mentioned before, RDG is designed to exploit the memory channel to provide a low latency (microseconds) connection between systems. We have upgraded OS from Tru64 5.1A to 5.1B on 12/5. This was done by HP CE. Now we are reviewing all sysconfigtab parameters. We have already followed oracle metalink guides as you mentioned. However, there are still some questions,

1. Is there any change regarding RDG between 5.1A to 5.1B?

2. How to monitor the memory channel performance between 2 nodes?

Sincerely thanks for your full support.

Eric

sysconfigtab 之 max_sessions???

watermelonyu — Thu, 09 Dec 2004 03:01:05 GMT

Hello Eric:

As I know, currently, thers is no specific

doc that has information for any change\

regarding RDG between 5.1A to 5.1B

If you want to monitor memory channel, you

can use the following commands:

# collect -i 120 -sn (because memory channel

has ip assigned, so you can use collect to

monitor it as network for packet counts)

# monitor (this can be downloaded from HP, as

follow:

ftp://ftp.digital.com/pub/DEC/monitor.alpha.tar.Z, please refer to the attachment for monitor

output.

# cfsstat -i 15 -S icssvc | tee -a \

/var/adm/ics_`date +%y%m%d`.log, then you

can use tail -f /var/adm/ics_`date +%y%m%d`.log

or vi it for viewing, you can use man cfsstat

for more information.

Best Regards

Danny Lin