bahman/scylla
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

docs: explain tuning for a node that is overcommitted at the hypervisor level

Browse Source 
Closes scylladb/scylladb#19589
This commit is contained in:
Avi Kivity
2024-07-02 21:36:19 +03:00
committed by Botond Dénes
parent 8809b99736
commit 69450780a7
1 changed files with 24 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
docs/getting-started/scylla-in-a-shared-environment.rst
View File

@@ -65,6 +65,30 @@ On Red Hat / CentOS edit ``/etc/sysconfig/scylla-server``, and add
If starting ScyllaDB from the command line, simply append
``--smp 2 --overprovisioned`` to your command line.
Overprovisioning at the Hypervisor level
----------------------------------------
Sometimes, you can dedicate the entire node to ScyllaDB, but the node itself
is running on a hypervisor that overcommits the CPU (this is not true for most
public cloud instance types). You can tune ScyllaDB to
this configuration by using the command line flags ``--idle-poll-time-us 0 --poll-aio 0``.
These settings tell ScyllaDB to never poll while waiting for a network or disk event,
and instead relinquish the CPU quickly. This lets the hypervisor schedule other
virtual machines, and in turns avoids penalizing the ScyllaDB node for overconsumption.
These settings are a subset of the `--overprovisioned` flag (omitting memory and affinity settings).
On Ubuntu edit ``/etc/default/scylla-server``, and add
``--idle-poll-time-us 0 --poll-aio 0`` to tune ScyllaDB for an overcommitted hypervisor.
On Red Hat / CentOS edit ``/etc/sysconfig/scylla-server``, and add
``--idle-poll-time-us 0 --poll-aio 0`` to tune ScyllaDB for an overcommitted hypervisor.
If starting ScyllaDB from the command line, simply append
``--idle-poll-time-us 0 --poll-aio 0`` to your command line.
Other Restrictions
------------------