Re: Jesper AC What I don't like
".....When he is proven wrong....." Que? How is showing that the p260 only has two IO slots and no onboard LAN ports being shown to be worng when it illustrates the glaring compromise of the p260 design? In an HA environment, when I will want two LAN mezz cards and two FC cards minimum for redundancy, your only answer is to double up on converged cards. Leaving no other slots for high-speed interconnects like Infiniband, not to mention the SAS option if Flex even had a SAS option. The hp Itanium blades all come with two onboard dual-port converged adapters, which means I have redundant LAN and SAN just by adding a second converged mezz card, leaving two slots for linking to a second SAN or a SAS switch, or for redundant Infiniband cards. You can pretend that not having redundancy is not an issue if you like, but I seem to remember it being quite key to HA clustering.
"......Sure 2 adapters with will address all four switches in the back. Try reading a manual....." Oh dear, back to RTFM for you! Two IO slots on the P260 means NO options, it is simply too few. Does IBM offer a mezz card that can magically do LAN, SAN and Infiniband? No. So if you have one converged card and one Infiniband card, where is the redundancy required for clustering? You have none. One card failure and your p260 is a dead duck.
But back to the manuals you mentioned. When the original BL860c came out I asked the hp guys in Vegas if they could design a half-height Itanium blade and then said they could, but it would be too much of compromise. The p260 simply proves that point. You mentioned manuals, well go look here (http://www.redbooks.ibm.com/abstracts/tips0880.html#locations) at the layout of the p260, note how the CPUs are at the front and their heatsinks cover all the airflow to the memory? Once again, IBM have designed a blade where the CPUs will cook the memory! And the disks, which hinge over the memory banks and are not hot-swappable (as they are on the hp Itanium blades). Yes, yet again, IBM have designed a blade where you have to take it out of operation and open it up to take out a failed disk - have they learnt nothing from their previous failures? The disks also limit the height of the memory and the airspace around it for cooling, meaning more pricey special low profile memory for the IBM Power blades.
But whilst we're on the disks, does the p260 have hardware RAID for the onboard disks as the hp blades do? Nope, software RAID only! This, however, does not seem to be a design compromise as just a lack of design, the full-height p460 also not having onboard RAID. So not only do you have to yank a blade if a disk fails, you have no RAID protection for the data on the internal disks unless you give up CPU cycles to software RAID.
So let's summarise - the p260 has no onboard adapters, no hardware RAID, no hot-swappable disks, not enough IO slots for real flexibility and redundancy at the same time, and will probably cook its expensive memory (along with those non-hot swappable disks) as soon as you start thrashing it. Yeah no compromise at all - NOT! But even funnier is IBM still haven't worked out how to make a big blade, even though the p-series rack servers have been modular for ages - four sockets is the best they can manage, whilst hp's blades have managed eight sockets as one hardware instance for years! That means whilst hp can run an eight-socket octo-core blade, with IBM if you want to scale that high you have to throw away the Flex chassis and go buy an IBM rack server and a whole lot of external switches.
Stop pretending that p260 is anything other than a compromise design to counter the Xeon BL460c Gen8, nothing more. Even the IBM literature describes it as a "compute node", exposing their intention for it to be HPC and nothing else. It simply does not have the HA feature set for enterprise UNIX clustering.