HPE 9000 and HPE e3000 Servers
cancel
Showing results for 
Search instead for 
Did you mean: 

rp8400 Questions

 
SOLVED
Go to solution

rp8400 Questions


I have a customer to whom our company is proposing a used rp8400. It will replace their aging N-class system. The IBM sales rep has been beating us up pretty badly, saying the rp8400 is "old technology", that it "can't recover from various failures", and that "it isn't expandable and won't give you enough growth potential". Worse, their N-class has been failing quite regularly and has been a disruption in their business. As such, HP has something of a bloody nose right now, which isn't helping. Worse, the HP Service Rep, while onsite fixing their N-class, told the customer that the N-class and the rp8400 were "basically the same" and he shouldn't expect more uptime from an rp8400 than his N-class.

Our company installed a used rp8400 internally (which I admin) for development late last year (also displacing an old N-class) and I have been very happy with the platform. Frankly, I'm just a developer and don't generally get involved in this sort of battle, but since I have direct experience with our rp8400 and the sale is very large, I got pulled in.

My questions are as follows:

o For those of you who use rp8400's and/or like systems (e.g. rp7410, first gen Superdome) that use the same chipset, what failure rates have you seen on non-disk parts? Did any of these failures crash the system?

o Has anyone had a Core I/O board fail? What happened to the system? Our customer has had their GSP fail several times and it has crashed their N box repeatedly.

o Have you ever had a CPU fail while the system was in operation? Did it take the machine down? HP's docs go to great length explaining that the rp8400 can dynamically deallocate CPU's when they start logging errors, but our customer has experienced multiple CPU failures on the N-class and each one has crashed the machine. IBM touts the p-Series as essentially "crash proof" from a CPU failure. I've sent the customer docs from IBM to dispute this, but he's really leaning towards the IBM solution. I'd like some real-world input to help allay his fears.

o What is the minimum number of nPars you can configure on this system with the expansion cabinet? (i.e. if I wanted to put 32 CPU's in a single nPar long-term, is that possible?) We're proposing a fully loaded rp8400 with 16 CPU's and 32GB of RAM in two nPars, but would like to know if we can grow those two nPars into the expansion cabinet at a later date, or, assuming they get that large, only have a single nPar plus a second rp8400. I've looked at the partitioning guide, but I didn't see anything obvious about the minimum number of nPars.

o Any other positives about the rp8400 from anyone that uses both HP and IBM in their shop?

Many thanks in advance for your help!
2 REPLIES 2
Solution

Re: rp8400 Questions

o For those of you who use rp8400's and/or like systems (e.g. rp7410, first gen Superdome) that use the same chipset, what failure rates have you seen on non-disk parts? Did any of these failures crash the system?

Excellent MTBF (Mean Time Between Failures) on most parts. I've had a few procs and memory sticks take a dump, but definatly enterprise class!

o Has anyone had a Core I/O board fail? What happened to the system? Our customer has had their GSP fail several times and it has crashed their N box repeatedly.

Yes. This was after the system was shipped across the Atlantic and noticed at poweron. I've played with HPs a LONG time (and N-Classes are good boxes) and it's the only time I've seen it (knock-on-wood).

o Have you ever had a CPU fail while the system was in operation? Did it take the machine down?

Yes, and yes. This depends alot on the way the CPU fails in both HP and IBM. IBMs have a higher sucess rate of noticing hardware failures and elegantly failing (in my OPINION).

o What is the minimum number of nPars you can configure on this system with the expansion cabinet?

You don't need any N-par (1 partition) as the minimum - 1 partition can access all 32 potential slots if you like.

If I wanted to put 32 CPU's in a single nPar long-term, is that possible?
Yes, but you might have to get the other GSP and Core I/O Card to access all four internal drives and all 16 internal slots.

We're proposing a fully loaded rp8400 with 16 CPU's and 32GB of RAM in two nPars, but would like to know if we can grow those two nPars into the expansion cabinet at a later date.
Yes, the SEU (Server Expansion Unit) can be assigned to NPars and even more versitle to VPars. This assignment can change with fairly simple maintenance depending on what cards (and remote devices) you are moving.

o Any other positives about the rp8400 from anyone that uses both HP and IBM in their shop?

I've been places where two rp8420's ran the 2+billion $ business for >3 years without problems. Those chassis started life as rp8400's then went through field upgrade to 8420's. I've seen over 20 N-classes run with minimal downtime for almsot 3 years (downtime usually b/c of crappy 4si raid cards).

I'm not sure of your setup, but I would try to do something like TWO rp-7440 (16 cores each) with some service guard (or Veritas clustering) between them. Again, that's a guess not knowing specifics.

IBM's do have some slick Partitioning compared to HP. But the prices are a bit higher too.

Hope this rant heps some.
Alzhy
Honored Contributor

Re: rp8400 Questions

o For those of you who use rp8400's and/or like systems (e.g. rp7410, first gen Superdome) that use the same chipset, what failure rates have you seen on non-disk parts? Did any of these failures crash the system?

Fairly "high" IMHO. This judging from an ecosystem of 5 rp8400/8420s(32-way each) and 1 SuperDome/32 (64-way). Expect the occasional Cell Board replacements due to CPU failure modules, CoreIO Boards (plenty). Did it crash the system? Absolutely! That is why - for Critical systems -- Clustering is a No-Brainer.

DOn't get me wrong though - these machines are good and reliable workhorses. Use them properly thoug if you want higher availability (clustering...)

o What is the minimum number of nPars you can configure on this system with the expansion cabinet? (i.e. if I wanted to put 32 CPU's in a single nPar long-term, is that possible?) We're proposing a fully loaded rp8400 with 16 CPU's and 32GB of RAM in two nPars, but would like to know if we can grow those two nPars into the expansion cabinet at a later date, or, assuming they get that large, only have a single nPar plus a second rp8400. I've looked at the partitioning guide, but I didn't see anything obvious about the minimum number of nPars.

rp84XX families can go up to 2 nPars (hard partitions). If you've deeper pockets - I'd suggest going with vPars as it offers you utmost flexibility and CPU (and memory with later releases) dynamism. With vPars you can go up to 16 vPars each independent from each other and running its own OS and HBA shares. CPU granularity is whole cpus though.

o Any other positives about the rp8400 from anyone that uses both HP and IBM in their shop?

None.
Hakuna Matata.