Re: SQL Node clarification

[Shaheeb Roshan]

I've found a few of my answers and a lot more questions!

As an FYI:
- adequate physical memory is only half the battle, you have to ensure that the operating system will allow you to allocate it to a single process. I'm using Slackware running a 2.4 kernel, and simply had to recompile it using HIGHMEM and it worked like a charm.

- Stewart has a neat little size estimation script called ndb_size.pl that will analyse a given db and report the size changes in 5.0 and 5.1. Because 5.0 uses fixed row lengths, and 5.1 isn't GA yet, we're stuck with a not too efficient size increase. Please note that this ndb_size.pl is *not* included the 5.0.19 or 5.0.18 max binaries. I had to find it as an attachment to a post on the forums.

Now if some can help me clarify the following:

- The docs mention you can run more than one ndbd process on a single machine, effectively running more than one storage node on a single machine. I understand the performance benefit of having each single process allocating less resources to itself. So instead of a single ndbd process sucking up 2 gigs, 2 processes can split it up 50/50. But when I tried to configure this, I got warnings that if the host for node group 0 went down, the whole cluster would go down.

my config snippet:
*********************************
[ndbd_default]
NoOfReplicas=2

[ndbd]
id=2
hostname=xx.xx.xx.20
datadir=/usr/local/var
[ndbd]
id=3
hostname=xx.xx.xx.20
datadir=/usr/local/var

[ndbd]
id=4
hostname=xx.xx.xx.21
datadir=/usr/local/var
[ndbd]
id=5
hostname=xx.xx.xx.21
datadir=/usr/local/var
*********************************


Should I change the datadir settings for each host? I mean, is the problem that both ndbds on xx.20 are pointing to the same datadir? If I make the change, will the cluster then produce 2 replicas, just among 4 data nodes?


Thank you for your patience!

Shaheeb R.