RP8400 SPOFs

[Asghar Ghori]

Does anyone know what the single points of failure are in the RP8400 servers.

I also need to have a diagram that shows the entire box hardware architecture.

Thanks

[Joshua Scott]

as for the architecture, see here:

http://docs.hp.com/hpux/hw/index.html#rp8400%20Server

The 'HP System Partitions Guide: Administration for nPartitions' has a diagrams on page 39, (on the PDF file) and may have some more throughout.

as far as SPOF, the Backplane is tha only thing I know of. The HA feature include:

N+1 hotswap cooling

Redundant and HotSwap power supplies

Cell Hot Plug

hot-plug disks

2N power inputs

Online memory page deallocation

ECC protected SyncDRAM memory

Full parity protection of data and address buses

On-Chip CPU cache w/ ECC protection

Memory "Chip spare", "Chip kill" like

CPU deallocation on failure

On-line addition and replacement of PCI I/O cards

Four Independant UltraSCSI buses to internal disks for mirroring across disks and controllers

Journal File System

Auto Reboot

On-line Diagnostics and system health Monitor

HTH.

-Josh

[Jeff Schussele]

Hi,

Backplane, I believe, in this case means the interconnect between the two cell boards.
I think another is the PCI interconnect between the 2 PCI busses (sp?).

Rgds,
Jeff

[Joshua Scott]

I beleive you're correct, jeff.

[Stefan Stechemesser]

The two halfs of the PCI Backplane are completely independent from each other (except they are on the same board and cannot be changed independendly). Each half has 8 seperate PCI busses connected to a system bus adapter which is connected to a cell controller on the cellboard. If one cell want to access IO connected to another cell, it has to connect via the cell to cell connections on the backplane. There is no direct PCI to PCI bridge or soomthing like this. If one half of the PCI cardcage fails completely (power is also independent) the partition has still the I/O of the other half (if configured redundant).

=> PCI is no SPOF

(OK, I forgot to say that the system crashes if on half of the PCI backplane fails, because HPUX does not like something like this, but after the reboot you can continue working without the 2nd half of your PCI cards ;-)

[Michael Steele_2]

Perhaps you should be looking at single points of failure in a nPar environment on a RP8400 as well. For I believe a single point of failure exists with boot disk controllers in a npar environment. Because boot disks can't be shared amongst npars. You run out of internal disk controllers quickly for this reason. For example, 2 partitions with one internal disk controller on either npar. Because boot disks can't be shared between npars this means the primary and alternate boot disks are on one controller. And unless you buy 2 additional disk controllers for a 2 npar configuration your internal boot disk controller becomes a single point of failure.

Now build a 4 npar environment. Eight boot disk controllers and 16 disks to cover all the primary and alterate boot disks for 4 npars.

So boot disks as well are SPOF's in a npar environment.

[Michael Steele_2]

Sorry, "...16 disks..." should be "...eight..."

[Joshua Scott]

Michael,
The rp8400 comes standard with 4 controllers, and only goes to 2 nPars. Thus you have redundancy on the controllers, if you have all 4 internal drives, and add mirrordisk (which I am assuming would be done if one is concerned at all about High Availability)

The only way the tp8400 goes to 4 npars is if the Server Expantion Unit(SEU) is added. the SEU is another enclosure with 4 more internal disks and 2 more Removable Media bays, as well as 16 more I/O slots and 2 more core I/O cards. Thus, even when going to 4 nPars, you still get redundant disks/disk controllers.

-Josh

[Michael Steele_2]

Gee, I guess.

This SPOF with disk controllers and boot disks has been around, is a known problem and is especially a problem in HA SG environments.

As you stated there are four SCSI controllers but the RP8400 can support 4 partitions. That's one controller per partition.

See page 86:

http://docs.hp.com/hpux/onlinedocs/hw/rp8400/en/documentation/cust.pdf

#############################################
############################################

Here a chart of the 4 standard nPar config. for an 8400, note 4 is standard except in HA environments:

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/5187-3603/5187-3603_top.html&con=/hpux/onlinedocs/5187-3603/00/00/28-con.html&toc=/hpux/onlinedocs/5187-3603/00/00/28-toc.html&searchterms=rp8400&queryid=20031009-084852

#############################################
############################################

Here is an 'ioscan' from an nPar server, not necessarily an 8400. Please note that all the internal boot disks are on same controller at SCSI targets 5 and 6, while the controller is on target 7. Also note that this is for SCSI bus priorities.
Target 7 has high priority and is always reserved for the SCSI controller. While targets 6 and 5, then next lowest priorities, are always reserved for boot disks.

12/0/6 ba Local PCI Bus Adapter (782)
12/0/6/0/0 ext_bus SCSI C87x Ultra Wide Differential
12/0/6/0/0.5 target
12/0/6/0/0.5.0 disk SEAGATE ST39173WC
12/0/6/0/0.6 target
12/0/6/0/0.6.0 disk SEAGATE ST39173WC
12/0/6/0/0.7 target
12/0/6/0/0.7.0 ctl Initiator

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/5187-3603/5187-3603_top.html&con=/hpux/onlinedocs/5187-3603/00/00/17-con.html&toc=/hpux/onlinedocs/5187-3603/00/00/17-toc.html&searchterms=controller&queryid=20031009-085141

#############################################
############################################

Other issues with SPOFs"

For high availability environments where SPOF's are critical, please note this quote from the URL below:

"...The PRI and HAA paths should be configured to reference disks that are connected to different cells,..."

#############################################
############################################
nPartitions Guide:

http://docs.hp.com/hpux/onlinedocs/hw/rp8400/en/documentation/spg.pdf

#############################################
############################################

As others have pointed out, SPOF at the 2 core I/O boards through power which flows through single PCI back plane.

[Joshua Scott]

Michael,

I will have to disagree with you. The rp8400 does not support 4 nPartitions. See here:

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/5187-3603/5187-3603_top.html&con=/hpux/onlinedocs/5187-3603/00/00/28-con.html&toc=/hpux/onlinedocs/5187-3603/00/00/28-toc.html&searchterms=rp8400&queryid=20031009-084852

Note at the bottom of this document it states:
"On rp8400 servers, each nPartition must include either cell 0 or cell 1 because these two cells are the servers' only core-capable cells"
A 'core-capable' cell is necessary for an nPartition, therefore with only 2 core-capable cells you cannot have more than 2 nPartitions.

-Josh