HP forges 32-socket Itanium iron
Beyond Superdome
Magic Quadrant for Enterprise Backup/Recovery
What's going on with Hewlett-Packard's Itanium-based Integrity server lineup? It has been over a year since HP got its first machines using Intel's "Tukwila" quad-core Itanium 9300 processors into Integrity machines, and the company has still not delivered machines spanning 32 and 64 sockets. This is the kind of big iron that key HP-UX accounts need.
IBM got its big, bad Power 795 AIX box out the door last fall, sporting up to 256 Power7 cores in a single system image, and the machine was one of the key factors driving up the company's Unix server sales by 19 per cent in the first quarter of this year. Oracle and Fujitsu are in super-slow-mode with the Sparc Enterprise M9000, sticking with the existing quad-core Sparc64-VII+ chips and able to field 256 cores in a 64-socket configuration. The Superdome 2 is stuck at 64 cores.
The lack of a high-end Integrity server running HP-UX is simultaneously a boon for Oracle, which wants to sell its own (or rather, rebranded Fujitsu) Sparc Enterprise M8000 and M9000 servers, as well as a bust because Oracle does in fact sell a lot of its database, middleware, and application software on HP-UX machines.
If HP can't deliver big Superdome 2 boxes, Oracle might as well bash the Itanium chip, as it has been doing. Oracle says that Intel and HP are not being honest about the future of Itanium, that Intel is focused on Xeon processors, and has therefore said that it will not develop future versions of its database, middleware, and application software for the Itanium platform. And when Oracle says that, what it is really saying is that it is stopping development on HP-UX.
Publicly, Intel has stuck to its party line, saying that it has two more Itanium generations coming down the pike, but has told its Itanium customers privately that it is considering its options for what happens after the "Kittson" Itaniums come to market in 2014 or so. The eight-core "Poulson" Itaniums, now confirmed for 2012, will plug into the current Integrity blade and rack servers, which were announced last April, and they are also plug-compatible with the Superdome 2 blade-ish SMP boxes, which currently come with 8 or 16 sockets in a single system image using the Tukwila Itanium processors.
The sx3000 chipset implements a crossbar interconnect that provides up to 32 links per blade enclosure; each enclosure can hold a maximum of eight blades. Kirk Bresniker, vice president and chief technologist for the Business Critical Systems division at HP (and an HP Fellow as well), tells El Reg that this Superdome 2 crossbar is based on HP's own electrical design and uses its own communication protocols, which have their heritage in the Convex supercomputers and the PA-RISC and Itanium Superdome machines.
The early Superdomes had their own custom cables to implement the interconnect, but with the SD2 machines, HP has switched to PCI-Express cabling, which saves money. In any event, these crossbar links can be used to lash multiple processor boards together and to link out to external I/O expansion enclosures. That crossbar is implemented in separate blades that go in a separate chassis on top of the enclosure for the processor blades, linking out to external I/O or adjacent crossbar blades feeding into other Superdome 2 blade enclosures.
The main differences between the SD2-8s and SD2-16s machines, as these servers are called, is the size of the nPar partitions that run on the boxes and external I/O expansion. Both are based on two-socket extended blades that have the sx3000 chipset from HP plugging into the QuickPath Interconnect (QPI) point-to-point links on the Tukwila chips to implement SMP clustering for server nodes.
These machines support the Itanium 9340 and 9350 processors, from 32GB to 2TB of DDR3 main memory, 24 PCI-Express 2.0 slots, and 32 10 Gigabit Ethernet ports. The SD2-8s has nPar hard partitions that can span as large as eight processors and 32 cores, has up to 48 external I/O ports that can be attached into the sx3000 chipset's crossbar. The SD2-16s can have an nPar hard partition that can extend across 16 sockets and 64 cores (that's the entire box) and has double the external PCI-Express peripheral slots, at 96.
The HP Integrity server roadmap from late 2009 had the Integrity blade servers - the two-socket BL860c i2, four-socket BL870c i2, and the eight-socket BL890c i2 - being launched in April 2010, with the SD2-16s coming in August 2010.
There was no mention of an SD2-8s in that late 2009 roadmap, but the SD2-32s, a 32-socket server sporting 128 cores and spanning two SD2 chassis, was due in the first half of 2011 and the rx2800 i2 rack server - mostly aimed at OpenVMS customers but also sporting Windows Server 2008 R2 - was due in the second half of 2010. As we know, HP kicked out the Integrity BL8X0c i2 blades, the rx2800 i2 rack, the SD2-8s, and SD2-16s machines in April 2010 and hasn't touched the product line since.
There have been some enhancements to the NonStop machines, which are also based on the Integrity platform, and at some point last year, HP tweaked the Superdome 2 online specs to let customers know the SD2-32s machine was coming and that the architecture could scale to 64 sockets and "from 8 to 256 cores and more".
Next page: Superdome silence
COMMENTS
OracleRdb ?
I suspect quite a few of them will run OracleRdb at this stage, just as they always have done - it is by far the best database available on OpenVMS.
Sadly, OracleRdb has been shot by Oracle as well - at this stage - so those folks at some REALLY big clients have several years work cut out for them converting their software to some "other database" on some "other OS". Not a pleasant thought I am sure, having tried it personally.
HP has wanted OpenVMS (and Tandem) to go away as quickly as possible ever since they acquired them, now their strategy is enabled by Oracle's poke in the eye. What luck for the HP execs, soon they can move on to more stock options and more interesting things that don't work. They haven't been able to kill OpenVMS themselves, so Larry lends a helping hand by decommissioning the 2 databases that a very, very large percentage of OpenVMS customers rely upon. Good to go Larry! No long term database upgrade path is about as clear a message as is possible for OpenVMS customers.
The "Alphacide" and the Itanium or nothing plan was always the death of OracleRdb (and NonStopSQL) - now I think the clock is counting down to the last decommissioning of what truly is the best piece of database engineering out there. Sadly OracleRdb requires OpenVMS, which despite its superiority in so many respects, has zero visibility in the IT journalists world, and even less in the potential customer world - thanks HP, we appreciate it sooooo much, not.
I earn my living in the SQLServer world now. The worst database I have used since Oracle V3.x (IIRC), and that, if you are old enough to remember, was particularly nasty
Sailing off into early retirement to let he unwilling, led by incompetent solve the problems created by the "vision" of Carley, and ego of Larry.
OpenVMS 8.4 supports a wide variety of Integrity hardware, not just BL8x0 i2 blades
Re: "You cannot slap OpenVMS 8.4, the most current release, on any Integrity machine, and in fact, it is only supported on the BL8X0 i2 blades" <
Actually, OpenVMS 8.4 supports a wide variety of Integrity machines -- see the complete list in the OpenVMS 8.4 Software Product Description at http://h71000.www7.hp.com/doc/spdalphaintegrity84.pdf
Lashing together?
Quote
Bresniker tells El Reg that OpenVMS 8.4 will span across all of the four blades in the most recent Tukwila blade servers, delivering 32-core scalability, and that customers who need more scalability than that can do VAXclusters lashing multiple blade boxes together
Sigh.
Come on El Reg. Just ask anyone who has used VMSClusters in production, they are not lashed together.
VAXClustering has been around since 1983. I witnessed some of the FCC RFI Testing of the kit in a place called 'The Bubble' near Marlboro, MA. I was testing a VAX 11/730 combo.
Since the release of the Alpha CPU the clustering was renamed VMSClusters.
If this is HP still saying VAX Cluster then I'd be rather worried if I were an HP customer.
Even nearly 30 years after its introduction VMSClustering gives a whole lot of features that those in the Unix/Linux world can only drool at in envy. A fully implemented OOTB cluster wide file system will be IMHO its lasting legacy.


IT infrastructure monitoring strategies
Requirements Checklist for Choosing a Cloud Backup and Recovery Service Provider
Data control in the cloud
Cloud based data management
Enabling efficient data center monitoring