Capacity Forecast Approach
a) Historical performance database (Statspack/AWR Report)
b) Looking the profile for (Correlate workload data with %CPU busy)
c) Cap models in EXCEL (Modeling)
Capacity Forecast Techniques
1.Capacity benchmarking
a)Load testing
2.Capacity trending
a)Linear trend analysis
b)Statistical approaches
3.Capacity modeling
a)Analytic modeling
b)Simulation modeling
Selecting Best Techniques
Load testing is necessary early in the deployment process
Statistical analysis is great to select a subset of systems for further analysis
Modeling is best when you need an accurate predication of the resources required to support a given workload
Why Queuing Theory
a) Queuing theory - certain types of lines can be described mathematically
b) All need to do is figure out the right formulas and plug in some numbers
Easy - but requires assumptions that may not hold in some situation
c) Simulation- we build a model of the system and play with it- this is more work but also more flexible - so the model can be more accurate
d) This theory permits the derivation and calculation of several performance measures including
The average waiting time in the queue or the system
The expected number waiting or receiving service
e) The probability of encountering the system in certain states, such as empty, full, having an available server or having to wait a certain time to be served
Queuing Theory Assumptions
“Infinite” number of processes
Arrival rate of l (Poisson distribution)
Unlimited queue lengths allowed
Single line queue
First-come-first-served (FCFS) queue priority
Only 1 server per process
Service rate of u (exponential distribution)
Mean length of service is s = 1/ u
Queuing Theory Formula
What these formula to tell us ?
Database Server Process arrive at the rate of 25 per minute. The server can serve one process in 2 seconds. (Assume Poisson arrival and exponential service rates).
A) What is the average utilization of the server ?
l = 25 processes /minute
u = 1 Process/ 2 Sec (1 min/ 60sec) = 30 processes / minute
B) What is the average number of processes in the line ?
C) What is the average number of processes in the system ?
D) What is average waiting time in line ?
E) What is average waiting time in the system ?
The arrival rate l
Transactions arrive into a computing system.
There are many statistics we can use to measure the arrival rate.
Common statistics from v$sysstat; logical reads, blocks changes, physical writes, user calls, logons, executes,user commit, and user rollbacks.
The Servers
CPU , Memory , Network and IO devices are servers.
Each transaction on the server consumes service time,S.
The service time is how long it takes a server to process a transaction.
The busyness of a server is called the utilization, U.
When a server gets above 70% utilized,transactions start to wait.
The Queues
When a transaction waits, it is placed into a queue.
Each queue has a length, L.
Each transaction is in the queue has a time W.
Performance decreases when a server gets busy and transactions queue.
This occurs at around 60% to 75%.
Response Time
Forecasting Steps
Customer requirement
Workload Data Collection
Utilization Data Collection
Identify the workload data for the driving utilization factor. Co-relate all the workload matrices with the Utilization matrices.
Calculate the extra CPU needed
Prepare the report
Data Collection
Oracle Enterprise Management
SAR files
Customized data collection Scripts
Identify the workload factor
Choose those Matrices as arrival rate whose correlation coefficient is highest among all the workload matrices.
Greater the number of sample for calculating correlation coefficient , the better the workload relation with the utilization.
The higher the correlation factor the better the accuracy.
Formulas
Real Life example
Analysis
Subscribe to:
Post Comments (Atom)
1 comment:
Probably I can say with this blog make, more some interesting topics.
Post a Comment