This will be ignored if \fBpre_read\fR is also specified for the
same job.
.TP
-.BI sync \fR=\fPbool
-Use synchronous I/O for buffered writes. For the majority of I/O engines,
-this means using O_SYNC. Default: false.
+.BI sync \fR=\fPstr
+Whether, and what type, of synchronous I/O to use for writes. The allowed
+values are:
+.RS
+.RS
+.TP
+.B none
+Do not use synchronous IO, the default.
+.TP
+.B 0
+Same as \fBnone\fR.
+.TP
+.B sync
+Use synchronous file IO. For the majority of I/O engines,
+this means using O_SYNC.
+.TP
+.B 1
+Same as \fBsync\fR.
+.TP
+.B dsync
+Use synchronous data IO. For the majority of I/O engines,
+this means using O_DSYNC.
+.PD
+.RE
+.RE
.TP
.BI iomem \fR=\fPstr "\fR,\fP mem" \fR=\fPstr
Fio can use various types of memory as the I/O unit buffer. The allowed
will perform I/O within the first 20GiB but exit when 5GiB have been
done. The opposite is also possible \-\- if \fBsize\fR is set to 20GiB,
and \fBio_size\fR is set to 40GiB, then fio will do 40GiB of I/O within
-the 0..20GiB region.
+the 0..20GiB region. Value can be set as percentage: \fBio_size\fR=N%.
+In this case \fBio_size\fR multiplies \fBsize\fR= value.
.TP
.BI filesize \fR=\fPirange(int)
Individual file sizes. May be a range, in which case fio will select sizes
single CPU at the desired rate. A job never finishes unless there is
at least one non-cpuio job.
.TP
-.B guasi
-The GUASI I/O engine is the Generic Userspace Asynchronous Syscall
-Interface approach to async I/O. See \fIhttp://www.xmailserver.org/guasi-lib.html\fR
-for more info on GUASI.
-.TP
.B rdma
The RDMA I/O engine supports both RDMA memory semantics
(RDMA_WRITE/RDMA_READ) and channel semantics (Send/Recv) for the
.TP
.B nbd
Synchronous read and write a Network Block Device (NBD).
+.TP
+.B libcufile
+I/O engine supporting libcufile synchronous access to nvidia-fs and a
+GPUDirect Storage-supported filesystem. This engine performs
+I/O without transferring buffers between user-space and the kernel,
+unless \fBverify\fR is set or \fBcuda_io\fR is \fBposix\fR. \fBiomem\fR must
+not be \fBcudamalloc\fR. This ioengine defines engine specific options.
.SS "I/O engine specific parameters"
In addition, there are some parameters which are only valid when a specific
\fBioengine\fR is in use. These are used identically to normal parameters,
\fInbd+unix:///?socket=/tmp/socket\fR
.TP
\fInbds://tlshost/exportname\fR
-
+.RE
+.RE
+.TP
+.BI (libcufile)gpu_dev_ids\fR=\fPstr
+Specify the GPU IDs to use with CUDA. This is a colon-separated list of int.
+GPUs are assigned to workers roundrobin. Default is 0.
+.TP
+.BI (libcufile)cuda_io\fR=\fPstr
+Specify the type of I/O to use with CUDA. This option
+takes the following values:
+.RS
+.RS
+.TP
+.B cufile (default)
+Use libcufile and nvidia-fs. This option performs I/O directly
+between a GPUDirect Storage filesystem and GPU buffers,
+avoiding use of a bounce buffer. If \fBverify\fR is set,
+cudaMemcpy is used to copy verification data between RAM and GPU(s).
+Verification data is copied from RAM to GPU before a write
+and from GPU to RAM after a read.
+\fBdirect\fR must be 1.
+.TP
+.BI posix
+Use POSIX to perform I/O with a RAM buffer, and use
+cudaMemcpy to transfer data between RAM and the GPU(s).
+Data is copied from GPU to RAM before a write and copied
+from RAM to GPU after a read. \fBverify\fR does not affect
+the use of cudaMemcpy.
+.RE
+.RE
.SS "I/O depth"
.TP
.BI iodepth \fR=\fPint
can increase latencies. The benefit is that fio can manage submission rates
independently of the device completion rates. This avoids skewed latency
reporting if I/O gets backed up on the device side (the coordinated omission
-problem).
+problem). Note that this option cannot reliably be used with async IO engines.
.SS "I/O rate"
.TP
.BI thinktime \fR=\fPtime