Groups keyboard shortcuts have been updated
Dismiss
See shortcuts

Two file I/O questions

26 views
Skip to first unread message

Niek Bouman

<niekbouman@gmail.com>
unread,
Feb 25, 2025, 8:50:39 AMFeb 25
to seastar-dev
In the seastar `/apps` folder there is an application `iotune` which performs some disk I/O measurements and then outputs the results to a file, which can then be provided to any seastar application with the `--io-properties-file` command-line argument.

- In case you do _not_ perform this tuning, can something be said about the performance penalty?

Another question:
The Seastar docs say that for good performance, it is advised to respect the `disk_read_max_length()` and `disk_write_max_length` parameters:
https://docs.seastar.io/master/classseastar_1_1file.html#a5ca4c63b1c643bd070c3c99840048511

My question is how to properly respect these limits when using the `dma_read` and `dma_write` overloads that take a `std::vector< iovec >`. Does the limit apply to each `iovec`-length separately, or should the _sum_ of the lengths of all `iovec`s in the vector not exceed this limit?

avi@scylladb.com

<avi@scylladb.com>
unread,
Feb 25, 2025, 1:51:21 PMFeb 25
to Niek Bouman, seastar-dev
On Tue, 2025-02-25 at 05:50 -0800, Niek Bouman wrote:
In the seastar `/apps` folder there is an application `iotune` which performs some disk I/O measurements and then outputs the results to a file, which can then be provided to any seastar application with the `--io-properties-file` command-line argument.

- In case you do _not_ perform this tuning, can something be said about the performance penalty?


In fact bandwidth can be better, but there will be no latency isolation between scheduling groups.



Another question:
The Seastar docs say that for good performance, it is advised to respect the `disk_read_max_length()` and `disk_write_max_length` parameters:
https://docs.seastar.io/master/classseastar_1_1file.html#a5ca4c63b1c643bd070c3c99840048511

My question is how to properly respect these limits when using the `dma_read` and `dma_write` overloads that take a `std::vector< iovec >`. Does the limit apply to each `iovec`-length separately, or should the _sum_ of the lengths of all `iovec`s in the vector not exceed this limit?


Sum of the lengths.

--
You received this message because you are subscribed to the Google Groups "seastar-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to seastar-dev...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/seastar-dev/a91325bf-464a-4d94-a56d-191134047db8n%40googlegroups.com.

Reply all
Reply to author
Forward
0 new messages