New server setup

So I have a Dell 6850 on the way with 32GB of ram and 6 hard drives

I am looking at doing a raid 10 for the DB and a raid 1 for the OS/tmp directory

Any other suggestions? Would a raid 5 with 6 drives be better then a raid 10 for mysql?

Dealing with a 2GB 5 million line item database where we delete about a 1/3 of lines each day and replace with new lines so theres a lot of inserts/deletes happening

Hi bbin,

I asked something similar on the mysql.com performance forum a little while ago and got a reply from Peter Z.

The link is here if you want to take a read:
http://forums.mysql.com/read.php?24,100894,100894#msg-100894

We’ve had the servers mentioned in that post up and running in production for quite a while now using a 4 disk RAID10 for the innodb datafiles (plus os etc) and a 2 disk RAID 1 just for the logfiles.

Even though this box does several hundreds of queries a sec (on 10’s of GBs of data) we’re not nearly hitting its limits, but what I can say is that the logfile disk is under constant write activity whereas the datafile disk is written to less often but in much bigger chunks. (which is as you’d expect really)

So even though it looks like the logfile disk is busier and should therefore be configured with more a more performant disk setup I still think that as disk load increases we’ll need a better disk setup for the datafiles as it’ll be random io (vs linear io on the logfile disks)

The size of database you’re dealing with is pretty small for that class of server so you shouldn’t have much in the way of problems if it’s setup reasonably well. I take it you have a battery backed controller cache to keep those writes/syncs nice and fast?

HTH,
Toasty

Anyone have any exp with memory raid setups? The server supports mirring and raid 5

Both would provide redundancy if memory failed but what really interests me is if raid 5 would increase memory performance?

When you say memory I take it you mean disk?

From the various bits and pieces I’ve read about this, RAID 10 will give you much better performance than RAID 5 with MySQL/InnoDB.

See this page for example: http://peter-zaitsev.livejournal.com/14415.html

Toasty

Ok so server is setup with a 6 disk raid 5 for the DB and a 2 disk raid 1 for OS/tmp

Question is this, by default the bin logs are stored in /var/lib/mysql with the database, would it be better to put those on the raid 1 to “move” the disk IO?

No I do mean memory, This server supports mirroring/redudancy at the memory level as well

Ooh, never come across this before. Is it intended to be some kind of replacement/improvement for ECC RAM?

Ok so server is setup with a 6 disk raid 5 for the DB and a 2 disk raid 1 for OS/tmp
Question is this, by default the bin logs are stored in /var/lib/mysql with the database, would it be better to put those on the raid 1 to “move” the disk IO?

Well like I said before we’ve split our main Innodb box as datafiles vs logfiles+os, and off the top of my head I can’t remember where we’ve pointed the tmp stuff to - probably the datafiles disk as I suspect it’s more random io… (we tend to use on-disk tmp tables very infrequently so it’s less important for us)

Originally I intended to do some detailed benchmarks to compare various setups but unfortunately I didn’t get the time I wanted to do this, so all I can really say is that the split of logs and data seems to be working really well so far.

Actually it’s just occurred to me that you might be using MyISAM not InnoDB, but either way you will probably want to focus on splitting datafiles and logs onto separate devices as your priority.

So to answer your question, I’d say “yes” in short )

Toasty