I understand that MySql Cluster's primary data storage is RAM to speed
up queries, but regarding the disk which is the best choice, SSD or
HDD?

Is MySql Cluster like other databases that accesses the disk
randomically or it is sequential since the data are queried in RAM?

Thanks
Flavio

Bishop <flavio.tobias (AT) gmail (DOT) com> wrote:
This depends. NDB does disk I/O for different reasons:

1. global checkpoints: the REDO log is flushed to disk. Logs are
pre-allocated. This is kind of sequential.

2. local checkpoints: the DataMemory is written to disk. This is
certainly sequential.

3. accessing disk data: although most data (anything indexed) is
stored in memory, you can also have columns stored on disk.
I/O is done in rather large chunks but still random.

4. backups: this is again sequential.


All except 3. is write-only for normal cluster operations (will only
be read at node (re)start or recovery). Especially 1. is critical and
requires low latency *and* high throughput.

The strength of SSD is low latency random reads. Exactly this is not
needed by NDB. Except for disk data and recovery of course. Then OTOH
SSD have (had?) weaknesses when it comes to heavy writes.

I haven't seen any NDB installation using SSD (nor laid my hands on it)
but IMHO the better (and much cheaper) choice are HDD.


XL