Blog sphinx test results

Performance indicators:

Performance	Max	AVG	Min
CPU usage		7%
Response time (seconds)	27 seconds	3 seconds	0.06 seconds
Searches per second (count)		1.8
Conclusion	According to the above data, the daily workload is 1.8/Second * 3600 seconds * 10 hours = 64 thousand searches Currently, the number of online searches per day is 6-, and the online hardware environment is far better than the test environment, Therefore, after the launch, each indicator is certainly better than the test result, but the disk has the risk of bottle diameter, as shown in the following analysis.

Test object:Use Sphinx for blog searchTest environment:Server 192.168.15.133 (configure dual-core 1.8 Memory 1G. The bandwidth is 100 m, which works normally on the desktop, 1 million data sources, and is stored on a disk)Test method:1. Five concurrent users access the page for 5 minutes. (Note: 10 concurrent users have been adopted, but the average response time exceeds 10 seconds, which is no longer within the user's tolerable range. Therefore, five concurrent users are used as the benchmark) 2. Randomly retrieve 10000 keywords for parameterized search to ensure different keywords for each search and avoid caching.Test Analysis:1. the CPU and memory used in the whole process are very low, and the load average is too high,
2. During the entire process, the disk is under high pressure (iostat-x). We can see that disk data interaction occupies 50% of the CPU. The disk is in a 99% busy state, and the average wait time for each I/O operation is 24 milliseconds, while the average service time for each I/O operation is 7 milliseconds and runs for a long time, too many disks in queue will become performance bottle paths