[Sunhelp] Fault tolerant system.

[Fletcher, Joe]

Hi,

My diagram didn't quite make it intact I see.


-----Original Message-----
From: Gregory Leblanc [mailto:GLeblanc at cu-portland.edu]
Sent: 21 September 2000 15:52
To: 'sunhelp at sunhelp.org'
Subject: RE: [Sunhelp] Fault tolerant system.


Disclaimer:  I don't have enough money to actually purchase a system like
this, but I've gone over it and diagramed it a couple of times, "just in
case".

Budget for tranquilizers and pain-killers if you do buy into this sort of
thing.
Alternatively use VMS! :-)



> -----Original Message-----
> From: Fletcher, Joe [mailto:joe.fletcher at Metapack.com]
> 
> I'm looking for some hints of putting together the following 
> hardware rig.
> 2x E420Rs, each with two dual SCSI adapters added and two 
> A1000 disk arrays.
> The system will be running RSF-1 "cluster" software. We will 
> have Veritas VM
> installed
> as well.  Initially I'd like confirmation of just how the 
> whole lot should
> be cabled together.
> This really needs a few diagrams to help explain the options but I'm
> assuming some people
> aren't usiing MIME enabled emailers. Lets call the servers S1 
> and S2. The
> arrays can be Ar1 and Ar2.
> Each array has two SCSI connectors, ports 1 and 2. Assume we 
> will be working
> with controllers in slot4 
> on each server. Each controller has ports A and B. Still with 
> me so far? 

Yeah, pretty straight forward.  

You'd think so but since it's not entirely obvious how the ports on the
A1000 behave
it's not quite. The trick is going to be making sure that if a failover
occurs, the standby server sees the same disk set as the primary. Likewise
if a disk set fails the standy set has
to have identical contents to the primary.


> What I think we need to do is connect 	S1 portA to Ar1 
> port1 and portB to
> Ar2 port2
> 					S2 portB to Ar1 port2 
> and portA to
> Ar2 port1

Sounds good to me.

So how about 	S1-A to Ar1-2 with S1-B to Ar2-1
		 	S2-A to Ar2-2 with S2-B to Ar1-1

There is a difference. It's further complicated by the fact that they want
to use the standby servers for Q/A testing so they will need a disk set of
their own to play on. 

> We then use RAID manager to carve up the disks within the 
> arrays (12 disks
> in each)
> and manage the resulting virtual disks through Veritas at 
> operating system
> level.
> 
> 	S1	S2		
> 	 |    \    / |
> 	 |      \/	 | 
> 	 |    /   \	 |
> 	Ar1	Ar2

Yeah, that's how I'd do cabling, as you're aiming for redundancy, so that if
any 1 piece fails, you still have the other to pick up.  

> Alternatives are to chain the two arrays together and single port each
> server thus:
> 
> S1----Ar1----Ar2 ----S2

I wouldn't go this way, as it's likely to hamper performance (the servers
don't have a bus dedicated to each array), and it takes a bit away from the
fault tolerance as well.  Later,
	Greg


This last bit is a stop gap if we can't get the planned config working
within the timescale
allowed by the client. Did I mention this has to be up and running by the
end of next week?