two hp rp8400

[David_492]

i have two Hp servers rp8400 and eva3000 array
and tape library MSL 5030 and we want to make the no single point of failure and clustering between servers and storage. our question that what is the consept of the hyberfabric cards which installed on each server and is it nessasry to have it in case we have OPS extension of RAC? does the HP PCI Hyperfabric2 adapter important in our case to make the OPS OPerates well ?

Thank you all

[sinhass]

See the link
http://www.docs.hp.com/hpux/pdf/T1859-90006.pdf

[James Anker]

David,

I assume your deploying 9i RAC on the DB level...?

The Hyperfabric is an HP proprietary standard of high speed interconnect between the nodes in a RAC cluster. RAC is very dependant on two things in the cluster, the first is latency & the second, bandwidth. Hyperfabric is very good in these areas and as such is the recommended interconnect You can use 1000 Base "T" LAN cards with Auto Port Aggregation S/W to achieve the same results although the speed will never be quite as fast as Hyper fabric. Other options you may wish to look at are Myrinet - they do a similar standard to Hyperfabric.

The bottom line is, Hyperfabric is fast and expensive, 1000 Base "T" is nearly as fast but a lot less expensive.

With Hyperfabric you need to use Hyperfabric switches (again proprietary HP) - these have a single point of failure (PSU) so make sure you get at least two. I have attached a diagram for you for one I have installed put in recently that similar to your setup.
You need 2 x Hperfabric or 1000 Base "T" cards per node as a minimum + one for heartbeat (see below).

You will need a copy of MC Service Guard with RAC Extensions per node and also a copy Veritas Cluster files system (will save you some major headaches further own the line)

Hope this helps you out david - the attached diagram should assist you to (assuming you have Visio!) - this is what you should be aiming for.

Kind regards,

James

[David_492]

Dear all thank you

Thank you alot james

Dear james i have another question for u .is it possible to connect the two HP rp8400 servers without Hyperfabric switches . incase they have hyperfabric2 adapters?

Thank you alot

[James Anker]

David,

Good question - I assume you mean point to point? If it were me I would say no for the following reasons:

1) MC Service Guard doesn't support switch less/point to point configs - MCSG is the cluster product used.

2) Switches enable the "scale out" instead of the "scale up" model to be utilised - this is one of the major benefits of RAC - i.e. more smaller nodes. The sweet spot for RAC is a 4 x node 4 x CPU cluster based on the RP7410 or RP5470. This provides the best trade of on price/performance & resilience.

3) I've not seen it as a supported config.

You also need to remember that in a two node RP8400 cluster one node loss = a 50% loss of cluster power however in a 4 x node cluster a node outage = a 25% loss of cluster power.
Also if you bring one node down for maintenance/patches etc you have one node left that has no resilience - loss of the second node whilst your working on the first could be embarrasing..

If your in the UK and need any more help drop me a line 07976 013265.

The Veritas product you need is CVM (Cluster Volume Manager)

Kindest regards,

James
HP9000 Tech Architect
SCC

[David_492]

Dear james

Again thank you alot

i enclose to you our connection attachemnt the digram includes the following :

1. Two Hp rp8400 servers
2. OPS RAC Extension software
3. LAN Backbone
4.EVA3000 , TWO Enclouser , each enclouser has 4 HDD 146 GB
5. Tape library for backup MSL5030
6. Two 16 port Fiber switches
7.currently there is no connection between the two servers i mean Hyperfabric2

I know i give u hard time
i got ur answers

but i still have some concerns

1. If we bring 4 pci Hyberfabric2 adapter two for each server can we make point to point connection from server to server as shown in the digram.becase u mentioned ops will not work unless we use hyber switches.

2. Dear james as u see the figure what u suggest?

3.And if i get hyberfabric cards does this mean i get active/active cluster with no single point of failure in this digram.

4. in case if there is no hyberfabric adapter can i use serial connection or scsi between the two servers.

Really i appreciate it

Thank you

Best Regards

















Connection Diagram.

[James Anker]

David,

The only alternative to Hyperfabric is 1000 Base "T" GBit LAN Cards (2 x per server) to use instead of the Hyperfabric cards - you need switches on all these configs for the interconnect and they should not be the LAN switches or it will run slowly.

SCSI between the Nodes is NOT supported - don't go there!

Follow my original diagram and you will be OK.

2C2D EVA Storage is fine - check the firmware is up to date.

(and assign some ITRC points!)

Cheers

James