topic Re: serve sizing in Operating System - HP-UX

serve sizing

Pradep — Tue, 15 Dec 2009 06:03:33 GMT

I have mcsg running on 2 nodes. 1 node runs application and other runs database. Both nodes are equally sized and can run aplication or database individually. it has been 3 years over without having parallel run of app and db on single node online with users connected.

i want to prepare for running them on single node as failover simulation.
can you please tell me what factors i need to look into performance wise to estimate if they can run together on single server.
like cpu/mem/swap utilization? run_queue_length? Any other factors i need to sum up for 2 nodes to get cumulative figure ?

Eg:-if both server have 4 cores each. and each of them 30% utilizaed..so when run parallel it might be 60% utilized on average. same for memory/swap. Is it a good way to get rough estimates ?

Please let me know what else i need to check for performance capacity check.

thanks.

Re: serve sizing

Johnson Punniyalingam — Tue, 15 Dec 2009 06:14:48 GMT

Hi Pradeep,

mcsg -> main concept not for load balancing its kind cluster setup to failover the nodes and applicatio to adoptive if in case of any "SOP"

Single point failure - like

patching,HW replacement, planned manintance
application failover testing. etc.

If both node are setup in same capacity or strong enough to handle each other d'not see much performance thing here.

as mentioned above post of your "Node 1 -> serves -> application and the other Node 2 runs as database. both have different purpose

Hope this Helps.

Regards,
Johnson

Re: serve sizing

Pradep — Tue, 15 Dec 2009 06:21:24 GMT

Thanks for the reply. lets see from another scenario. suppose i have 2 servers and want to consolidate into 1. what factors i need to look into.

Re: serve sizing

Johnson Punniyalingam — Tue, 15 Dec 2009 06:28:20 GMT

>>>Thanks for the reply. lets see from another scenario. suppose i have 2 servers and want to consolidate into 1. what factors i need to look into. <<<<

Are you looking for load balancing some thing like that ?

>>suppose i have 2 servers and want to consolidate into 1<<

Consolidate into 1>> in what ways ?

Re: serve sizing

Pradep — Tue, 15 Dec 2009 06:37:04 GMT

I want to decommission 2 physical servers and moved ther app/database to another new 1 physical server. No load balancing, only 1 server.

Re: serve sizing

Johnson Punniyalingam — Tue, 15 Dec 2009 06:52:21 GMT

I want to decommission 2 physical servers and moved ther app/database to another new 1 physical server. No load balancing, only 1 server.

>>New Server<< Do you have new server already installed ? are you planning to buy ?

Re: serve sizing

Pradep — Tue, 15 Dec 2009 07:27:04 GMT

not buy yet.

Re: serve sizing

Johnson Punniyalingam — Tue, 15 Dec 2009 07:36:02 GMT

check with "HP Singapore Sales Support Team"

let them know what "Application" Database" you are using on "old 2 decommission" servers.

They can do technical review of your "Set-up" advice accordingly

Re: serve sizing

Pradep — Tue, 15 Dec 2009 07:43:18 GMT

why only HP can do sizing ? even when i have running example of servers ?

Re: serve sizing

Johnson Punniyalingam — Tue, 15 Dec 2009 07:54:07 GMT

why only HP can do sizing ? >>>>

than what's value for people who do "System Architect" -> Just thought,

?>even when i have running example of servers?<< ?

how about any New feature, will much benfiet or boost your system performance of the Servers /applications/ Database.

Re: serve sizing

Michael Steele_2 — Tue, 15 Dec 2009 08:00:54 GMT

Hi

This is going to be a consumed resources issue. Run this command on both boxes to isolate your biggest cpu and virtual memory consumers:

UNIX96=1 ps -e -o vsz,pid,ppid,state,args | sort -rn | head -16

UNIX96=1 ps -e -o pcpu,pid,ppid,state,args | sort -rn | head -16

Combine the top resource consumers. Use your own judgment about how many to add together since each box will be different. If the top 20 are all grouped together then take the top 20.

Run this command on both to isolate your shared memory consumption:

ipcs -moba

Add up the SEGSZ column. Combine the total from both nodes.

Get your physical constraints: # of cpus. SGA. Physical memory.

Disk I/O use sar -d 5 5 and look at the %busy, avwait and avserv columns, avque, rw/sec, queue length. Question: Are any of the disks shared between both applications on both nodes? I would think not since one is an application server and other a database server. The database is probably hit its own disks, as is the application. If mutually exclusive then I would surmise that a disk I/O problem is unlikely. However, if you are combining disks, then you can add the rw/sec, queue length and $busy columns.

Approach this like any other performance trouble shooting problem, and determine how to combine the results for a projection.

Re: serve sizing

Michael Steele_2 — Tue, 15 Dec 2009 08:03:53 GMT

When finished, please remember to assign 0 to 10 points to all responses and also close the thread. Sometimes, when a thread is left open, years from now its responded too and gets several annoying comments from people. Thanks in advance!

Re: serve sizing

Pradep — Tue, 15 Dec 2009 09:47:39 GMT

Hi Michael,
i get this error

$ ps -e -o vsz,pid,ppid,state,args | sort -rn | head -16
ps: illegal option -- o
usage: ps [-edaxzflP] [-u ulist] [-g glist] [-p plist] [-t tlist] [-R prmgroup] [-Z psetidlist]
$ UNIX96=1 ps -e -o vsz,pid,ppid,state,args | sort -rn | head -16
ps: illegal option -- o

Re: serve sizing

Michael Steele_2 — Tue, 15 Dec 2009 10:57:57 GMT

Sorry, my bad.

UNIX95=1 ps -ef -o

Re: serve sizing

Pradep — Mon, 21 Dec 2009 07:04:23 GMT

Hi Michael,
I already have OV perf mgr installation. and i used GBL_CPU, GBL_MEM, GBL_SWAP, RUN_QUEUE_LENGTH for sum up. Can you pls tell me which other params of Perf mgr i need to addup..because it is easy to get historic data from it..like 1 yr old.

For virtual memory Swapinfo output is ok ?
# swapinfo
Kb Kb Kb PCT START/ Kb
TYPE AVAIL USED FREE USED LIMIT RESERVE PRI NAME
memory 5953088 1110452 4842636 19%

How do the relate adding SEGSZ value to kernel param SHm_segsz..are they is same format..like adding all SEGSZ together should be less that kernel value ?
. Combine the total from both nodes.

Get your physical constraints: # of cpus. SGA. Physical memory.
>> what is SGA ?

Disks are seperate in app and DB servers. so should not be a problem. Can i use any OV Perf param for checking disk util ?

Re: serve sizing

Michael Steele_2 — Mon, 21 Dec 2009 13:09:58 GMT

The SGA is Oracles reserved database memory. It should be as much as 90% of your physical memory.

You add the shared memory up on both nodes and sum them together. Shared memory is another constraint.

If you have a year of historical data then you're in good shape, but you'll have to get the MeasureWare dictionary out to id what your metrics are. I don't know them.

Re: serve sizing

Dennis Handly — Mon, 21 Dec 2009 16:17:06 GMT

>For virtual memory swapinfo output is ok?
# swapinfo

Please provide the "swapinfo -tam" output.

Re: serve sizing

Pradep — Wed, 23 Dec 2009 08:22:30 GMT

thanks for giving the approach for solution.