23.07.2014 Views

Lustre 1.6 Operations Manual

Lustre 1.6 Operations Manual

Lustre 1.6 Operations Manual

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

Parameter<br />

--regioncount_low=RL<br />

--regioncount_high=RH<br />

--regioncount_inc=P<br />

--regionnoise=k<br />

--regionsize=S[,S2,S3...]<br />

--regionsize_low=RL<br />

--regionsize_high=RH<br />

--regionsize_inc=P<br />

--threadcount=T[,T2,T3...]<br />

--threadcount_low=TL<br />

--threadcount_high=TH<br />

--threadcount_inc=TP<br />

--threaddelay=ms<br />

--fpp<br />

--verify-V=timestamp<br />

[,timestamp2,timestamp3]<br />

|--verify|-V<br />

Description<br />

Generate runs with a range of region counts starting at TL,<br />

increasing P% until the thread count exceeds RH. Each of<br />

these arguments is exclusive with the regioncount argument.<br />

When generating the next I/O task, do not select the next<br />

chunk in the next stream, but shift a random number with a<br />

maximum noise of shifting k regions ahead. The run will<br />

complete when all regions are fully written or read. This<br />

merely introduces a randomization of the ordering.<br />

The argument is a byte specifier or a list of byte specifiers.<br />

During the run(s), write S bytes to each region.<br />

The arguments are byte specifiers. Generate runs with a range<br />

of region sizes starting at TL, increasing P% until the region<br />

size exceeds RH. Each argument is exclusive with the<br />

regionsize argument.<br />

PIOS runs with T threads performing I/O. A sequence of<br />

values may be given.<br />

Generate runs with a range of thread counts starting at TL,<br />

increasing TP% until the thread count exceeds TH. Each of<br />

these arguments is exclusive with the threadcount argument.<br />

A random amount of noise not exceeding ms is inserted<br />

between the time that a thread identifies as the next chunk it<br />

needs to read or write and the time it starts the I/O.<br />

Where threads write to files:<br />

• fpp indicates files per process behavior where threads write<br />

to multiple files.<br />

• sff indicates single shared files where all threads write to<br />

the same file.<br />

Verify a written file or set of files. A single timestamp or<br />

sequence of timestamps can be given for each run,<br />

respectively. If no argument is passed, the verification is done<br />

from timestamps read from the first location of files<br />

previously written in the test. If sequence is given, then each<br />

run verifies the timestamp accordingly. If a single timestamp<br />

is given, then it is verified with all files written.<br />

Chapter 19 <strong>Lustre</strong> I/O Kit 19-15

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!