The Connect Test - PasaLab/tachyon-perf GitHub Wiki

##Introduction This benchmark is to test the metadata operation performance of Tachyon. It will setup a set of TachyonFS and request metadata operations to TachyonMaster concurrently.

When running, it starts X threads on each node, and each thread holds Y Tachyon clients. Each client executes the metadata operations sequently and totally there are Z operations. The X, Y and Z are user-defined.

For example, X = 4, Y = 3 and Z = 10000. There will be 4 threads, and we take a look at one of them. 'Thread-0' holds 3 clients, 'tfs-0', 'tfs-1' and 'tfs-2'. And the work process is:

'tfs-0' mkdir
'tfs-1' getFileId
'tfs-2' rename
'tfs-0' delete
'tfs-1' mkdir
'tfs-2' getFileId
...
(End when totally executed 10000 operations)

Each node and each thread works in their own workspace so there is no cross operation.

##Configuration The configurations of this benchmark is in conf/testSuite/Connect.xml.

##Result Demo This benchmark is supported to generate a total report by the tool TachyonPerfCollector. After finished a Connect Test, you can generate a total report with the command ./bin/tachyon-perf-collect Connect. Here is a demo of the total report:

Connect Test - ID : 1409646395767
Finished Successfully
********** System Configuratiom **********
NodeName	Cores
localhost	16
Total	16
********** Connect Test Settings **********
tachyon.perf.tfs.address	tachyon://localhost:19998
clients.per.thread	4
ops.per.thread	20000
threads.num	4
********** Metadata Operation Performace **********
NodeName	MetadataOperationPerformance(ops/sec)
localhost	15848.627
Total	15848.627
********** Node Details **********
localhost's metadata operation performance(ops/sec) for each threads:
	[ 3920.8 ][ 3969.829 ][ 3997.6016 ][ 3960.396 ]

In the total report, it will list the related system environment, the configurations and the performance result. In the Connect Test, we focus on the metadata operation performance and we show the result in ops/sec.