What is the best database setup with this server?

pdj

Running Navision 3.70B with a Native DB on 40GB.
I have new server with 6 physical disks. They are going to be RAID1 for protection, so I have 3 mirrored disks to work with. It has enough ram and isnt' doing anything else than being a NAV server.

Option 1:
Instal OS and NAV on Pair1 and use Pair2 and Pair3 for database parts.
Pro: DB drive is not used for OS
Con: Only 2 db parts and OS+NAV drive will have a very low load.

Option 2:
Create a 15GB RAID0+1 (stripe+mirror) partition across all 3 mirrored discs for OS+NAV, and 3 partitions running 2 disc RAID1 (mirror) so I can divide the database in 3 parts.
Pro: All disks have equal load
Con: OS+NAV on same discs as database.

I guess if I had 20 discs, then I should go for option 1 because the difference between 9 and 10 db parts is rather small.
And if I only had 4 discs I should take option 2.

But I have 6 discs :-) I'm for option 2, but a colleague of mine is considering option 1...

What do YOU think?

DenSter

I'd go with option 1. Keep your OS and programs separate from your DB's. That's a bummer if you have enormous space on those disks.

Savatage

We use your option 1 also..no problems

OS on seperate disk.

kine

I am voting for 1 too... It is better, you can keep your DB files of same size, it is extendable etc.

kriki

Or another solution:
diskpair 1: OS + DB-file of size X
diskpair 2: DB-file of size Y [EDIT:fixed the smiley that shouldn't be there]
diskpair 3: DB-file of size Y [EDIT:fixed the smiley that shouldn't be there]

and size Y > size X. How much? When running the system, you can run a form on "Database File" and check how the disks are used. And later adapt the sizes on the load of the disks.
Like this you use 3 diskpairs (3 databasefiles) but you put less DB-load on the first one because of the OS on it.

We have a client working like that with Navision 3.10 with a DB of about 5GB. Server of 512MB memory and it is going very fast.

PrebenRasmussen

My normal recommendations is to keep OS and DB apart.

Depends on the situation though. If you can promise me 100% that the server will never be used for anything else but a single Navision database server...

THEN I don't see a problem why you should not be able to place one extra database part on the first RAID1 also.

This will give you a theoretical performance gain of (3-2)/2 = 50% from using 3 instead of only 2 database files.

bbrown

The direct answer to your question is use Option 1.

Create the 2 database part the same size.
Grow them by even amounts.
Avoid growing by small amounts (percentage added)
Do not place anything else on those drives.
Optimize database on a regular schedule

Comments:

I think the real solution would be to add extra drives. Depending on usage needs (users and trannsaction types) I don't think 2 database parts will provide satisfactory performance on a 40 GB database.

Since the size of the database was known at the time of the server purchase, why was the server not purchased with more drives? Were you involved in the server purchase, or did some one just hand you the server to use? Is budget an issue? The cost of a few extra drives will be minor when compared with the potential cost of loss production when the system cannot perform.

Options:

1. Work with the client to upgrade the server configuration.

2. Give the client a choice between performance or fault-tolerance. Adding fault-tolerance to a low-end server can represent a sizable percentage of the total cost. If the client chooses performace then configure as RAID 0 and create 4 database parts.

pdj

Thanks for all your feedback - we will take them into consideration when creating the db :-) I'm still not completely convinced about opt2 is that bad. :-)

>kriki
Your suggestion about having different size db parts is also interesting. I have always heard that this was a NO-WAY solution, but I see you point. Anybody else experimenting in this area?

>PrebenRasmussen
I guess you stoke a nerve... Who can ever primise that the server in time wont be used for this and that... This could actually be the argument that convinces me to use opt 1 afterall.

>bbrown
Both budget and physical space (expanding existing rack) influences the purchase. So I was basically "given" the configuration, and simply told to get the best out of it. Well, the first configuration I was given was only with 4 discs, but I got them convinced to buy 6 right away (which is max for the server), and then buy Win2003 server Std. instead of Enterprise and only a single cpu instead of dual.
Selection between performance and fault-tolerance? The performance should be VERY bad before I would ask that question...

shogan@mila.ie

I have a similar setup, 6 disks, all 1+0, 3 arrays, first is OS, DB split into two parts, one RAID controller - Smart 642., using 65% of 12GB database.
Commit Cache = yes.

Question is: for the logical arrays, what stripe size should I use?

I have 2GB, DBMS = 700000, about 15 users, typically with two companies open at the same time on their PC.

Server not used for anything else.

I have tried <256kb on the DB drives, but I keep getting "Internal Error 100 in Module 19". So I am leaving it at 256kb, with the Logical Drive Array Accelerator Settings set to on for all logical drives. But is this OK?

kriki

I don't understand:
Do you have 6 disks in RAID10 or do you have 3 RAID1 pairs?

Savatage

shogan@mila.ie wrote:

"Internal Error 100 in Module 19"

Just to be clear, are you sure it's not Module 11?

shogan@mila.ie

Yes you are right - it should be "Internal Error 100 in Module 11"...sorry -

on our old server we keep getting "Internal Error 1355 in Module 19" on almost all clients...

I was very tired yesterday...