Trace file

[ATCP]

Hi,

I am trying to use your traces in my own experiments, but cannot understand information inside the trace.
Could you kindly briefly describe the fields in the files by taking some examples, like the fields in 0-disk-ec2-54-197-111-1.compute-1.amazonaws.com.blktrace.0.txt and 0-mem-ec2-54-197-111-1.compute-1.amazonaws.com.merged

Thanks and looking forward to your response

Ke

[pxgao]

Hi Ke,

   The schema is "batch_id, timestamp, start memory address (+ write, - read), length of request". memory address and length of request are in 4kb pages.

[ATCP]

Hi, Peter,

Thanks for your quick response!
is the timestamp in milliseconds? and the length of the request is the number of pages to write/read, am I correct? Thanks

Regards,
Ke

[shinyehtsai]

Hello Peter,

I grab one line of the trace from 0-disk-ec2-54-197-111-1.compute-1.amazonaws.com.blktrace.0.txt

202,16 7 50007 675.722691803 661 D W 95839968 + 80 [kswapd0]

There are around 11 columns.
But you only mention 5 columns,

batch_id, timestamp, start memory address (+ write, - read), length of request,

which are not enough to illustrate the trace itself.
Would you mind to provide some more contents?

Thanks.

[pxgao]

Hi, Peter,

Thanks for your quick response!
is the timestamp in milliseconds? and the length of the request is the number of pages to write/read, am I correct? Thanks

Regards,
Ke

Timestamp is in microseconds. Length of the request is the number of pages to write/read.

[pxgao]

Hello Peter,

I grab one line of the trace from 0-disk-ec2-54-197-111-1.compute-1.amazonaws.com.blktrace.0.txt

202,16 7 50007 675.722691803 661 D W 95839968 + 80 [kswapd0]

There are around 11 columns.
But you only mention 5 columns,

batch_id, timestamp, start memory address (+ write, - read), length of request,

which are not enough to illustrate the trace itself.
Would you mind to provide some more contents?

Thanks.

Hi disk trace is generated from blktrace and formatted by blkparse (https://linux.die.net/man/1/blkparse)

[ATCP]

Hi, Peter,

Could you kindly explain the experiment settings when collecting the trace, e.g. like what is link bandwidth and network delay, when you were collecting 0-mem-ec2-54-197-111-1.compute-1.amazonaws.com.merged

Thanks

[pxgao]

I am not sure about the following because it is a long time ago, but i think the trace is collected under 40G/5us

[ATCP]

Thanks, what about the local memory ratio you were collecting