Bxp Infrastructure Capacity
From All n One's bxp software Wixi
- 1 Methodology Used
- 2 Testing Profiles
- 3 Potential worked models
1 Methodology Used
Please note that this planning was confirmed in February 2016. In March 2016 bxp Release 10 saw a significant increase in server performance. bxp has not been tested against the improved server capacities but will exceed all figures currently stated. Bxp_Release_10
Capacity planning is essential to ensure that performance of the solution is effective for all scenarios. Given the extremely heterogeneous nature of the bxp environment it becomes “best guess” for items that are being prototyped.
There are formal approaches which can be used. Microsoft provide one potential approach in: http://msdn.microsoft.com/en-us/library/ff650695.aspx
Transaction Cost Analysis
- Step 1. Compile a User Profile
- Step 2. Execute Discrete Tests
- Step 3. Measure the Cost of Each Operation
- Step 4. Calculate the Cost of an Average User Profile
- Step 5. Calculate Site Capacity
- Step 6. Verify Site Capacity
- Step 1. Collect Performance Data
- Step 2. Query the Existing Historical Data
- Step 3. Analyze the Historical Performance Data
- Step 4. Predict Future Requirements
1.1 Existing Infrastructure Capacities
It is important to know what kit is available and identify the bottlenecks at different points.
1.1.1 Network Diagram
Supplied on application via Business Development manager
1.1.2 Core Network
1Gbps / 1Gbps uncontended.
1.1.3 Cisco ASA 5510
Table 2. Cisco ASA 5510 Adaptive Security Appliance Platform Capabilities and Capacities
- Up to 300 Mbps
Maximum Firewall and IPS Throughput
- Up to 150 Mbps with AIP SSM-10
- Up to 300 Mbps with AIP SSM-20
VPN Throughput Up to 170 Mbps Concurrent Sessions 50,000; 130,0001
Performance numbers tested and validated with Cisco ASA Software Release 7.2.
1.1.4 Barracuda 340
- 1 Gbps Maximum Throughput
- 35 Real Servers Supported
- 1U Mini Rackmount Chassis
- 1.2 Amps Input Current
- 2 x Gigabit Ethernet Connections
- Layer 4 Load Balancing
- Direct Server Return
- 500 TPS SSL Offloading/Acceleration
- High Availability
- Layer 7 Load Balancing
- Cookie Persistence
- SIP Call ID Persistence
- SSL Offloading
- Content Routing
Supplied on application via Business Development manager
2 Testing Profiles
The user profile within bxp varies from client to client and system to system, making creating an example user profile an almost impossible task. So instead we have three possible ways of reviewing the capacity.
2.1 Resource usage
With Classic ASP, a spike in the CPU indicates that the server received and processed a request as quickly as possible. This renders CPU monitoring unhelpful, except to identify protracted processing spikes which can be coding errors, extensive reports or temporary system usage spikes. RAM usage and Hard drive usage are measured but frequently and dynamically change depending on operational situation and time of the month.
2.2 Availability sampling
All n One will get immediate alerts from AWS if the servers reach any kind of storage or usage capacity that could cause a slowdown or make the service unavailable. The service is monitored 24/7 and any notification of a fault affecting availability will be issued to bxp immediately. This doesn’t really provide capacity planning but to state that the service was available.
2.3 Speed testing
The most effective capacity testing we have been able to use to date is to profile length of time from browser “request” to the finished response rendering. This testing has been facilitated by the extensive enhancements available in Internet Explorer 11 User Responsiveness capability.
Using timing and detailed breakdown of where a request / response test is performed the different aspects of the test must be individually measured. However overall experience is the measurement and provides the most globally useful testing.
2.4 Elements of the capacity test
When performing the test, we need to ensure consistency. For this certain elements may be ruled in or out. The key timing aspects are:
- Time of day
- Day of week
- Date in month
- Month of year
- Known project events
- Client machine used specifications
- Internet connection specification
- Load on internet connection / dedication of internet connection
- Running applications
- Interim connections (communications devices / hops between testing machine and bxp)
- HTML function type (page elements: text, images, video)
- Database functions required (Select, Update )
- Streaming elements (video, audio, conference, Citrix connection)
So with so many variables, devices and in flux components, simple stop watch testing is the most effective form of testing identified by All n One to date.
3 Potential worked models
3.1 Default Operation
All n One are sometimes requested by IT departments as to our capacity.
- Daily monitored stats put the operational statistics at 2 to 3% CPU usage on the web servers.
- < 3% average usage on the database server
- RAM file size constantly increases so is on an auto reset to ensure consistency.
- Hard drive space is monitored daily to ensure capacity is always available.
- Database speeds are reviewed using Percona SlowLog for systems analysis and Db query improvement.
Live operational displays supplied on application via Business Development manager.
3.2 Worked Example 1 - External site usage and uploading
3.2.1 File submission
Documents are usually transmitted to us from an interim server rather than direct file upload. So the capacity of transfer with the given infrastructure would be
Remote server upload speed: Unknown, assume unlimited.
bxp server accept speed: Bottleneck = Firewall 300Mbps
File size: 500Kb
Result: bxp at this level theoretically accepts 600 documents in 1 second with no other traffic or activity.
3.2.2 Information submission
A page request size varies depending on circumstance, however 500 bytes would be conservative estimate
This means bxp can handle 600,000 requests per second for pages simultaneously.
The returning page size varies greatly as well
Using the BEAPI, XML responses are sent. The XML with some data would not typically be much bigger. With a conservative guesstimate of 2Kb, we can guesstimate that bxp can supply 150,000 responses per second.