Magic Quadrant for Enterprise Backup/Recovery
California competition
IBM had not only get the 8 GB DIMMs out the door, but better get 16 GB DIMMs out there if it wants the HS22 to compete with California. At 192 GB of main memory, IBM can make a compelling case. Of course, Cisco is expected to use cheaper and lower capacity DDR3 DIMMs and just put more of them on the blade, tricking the Intel chipsets into thinking it is addressing less memory when it isn't thanks to the ASIC Cisco has designed. Even as IBM and other blade server makers adopt denser DDR3 main memory, Cisco's memory extension ASIC uses memory economics in Cisco's favor - provided that ASIC works well and doesn't cost much to make.
Speaking of virtualization, the HS22 blade server from IBM has an internal flash drive port, which can be outfitted with a 2 GB flash drive configured with VMware's ESXi skinny hypervisor. The blade has two hot-swap 2.5-inch disk bays, which can be populated with SAS drives (73 GB, 147 GB, and 300 GB at 10K RPM and 73 GB and 147 GB at 15K RPM), with SATA drives (300 GB at 10K RPM), or an SSD (weighing in at 31.4 GB).
The HS22 blade has two PCI-Express x8 slots, and has a two-port Gigabit Ethernet link standard, with up to 12 ports per blade possible with expansion cards. The HS22 also has expansion cards to deliver 10 Gigabit Ethernet and quad-data rate (40 Gb/sec) InfiniBand links. iSCSI, 8 Gb/sec Fibre Channel, and 6 Gb/sec SAS connectivity are available to reach out to disk arrays, too.
In terms of operating systems, the HS22 blade server supports Windows Server 2003 R1 and R2 as well as Windows Server 2008; Red Hat Enterprise Linux 4.7 and 5.3; SUSE Linux Enterprise Server 10; and Solaris 10 Unix. Only VMware's current ESX Server 3.5 and ESXi 3.5 hypervisors are supported on this blade.
IBM is deploying the Nehalem EP processors in a variety of its iDataplex custom-designed servers for hyperscale data centers. With this launch, the Nehalem chips are being deployed in four different configurations: a compute node, an I/O node, and two storage nodes.
Like the other IBM machines, the iDataplex nodes are based on Intel's 5520 chipset. IBM is supporting the 2.26 GHz L5520 (with a 60 watt TDP), the 2 GHz E5504, 2.26 GHz E5520, and 2.53 GHz E5540 (80 watt TDP), and the 2.67 GHz X5550, 2.86 GHz X5560, and 2.93 GHz X5570 (all 95 watt TDP) Nehalem chips in these systems. The iDataplex motherboard has 16 DDR3 DIMM slots, which support from 1GB to 4 GB DIMMs right now, for a maximum of 64 GB of main memory (the same 800 MHz, 1.07 GHz, and 1.33 GHz speeds as in the HS22 blade server). When 8 GB DIMMs are available in the third quarter, iDataplex customers will be able to plunk 128 GB of memory into these nodes.
iDataplex (Click to Enlarge)
The 2U iDataplex compute node puts two of these Nehalem servers stacked atop each other into a single chassis, each with two 2.5-inch SAS drives, two Gigabit Ethernet ports, and a single PCI-Express 2.0 x16 slot. The I/O server is a single server inside the same chassis, but it has two 3.5-inch SAS or SATA drives plus three PCI-Express 2.0 x16 slots.
The 2U storage server variant of the iDataplex node has five 3.5-inch drives and a single PCI-Express 2.0 x16 slot. For customers who need an even fatter storage node, IBM takes the bottom half of the 2U storage server and stretches the box to 3Us in height, and then puts in another dozen 3.5-inch hot-swap SAS or SATA drives (for a total of 13), including that single PCI-Express 2.0 x16 peripheral slot. Any of these iDataplex nodes can support the 31.4 GB SSD sold with the HS22 blade server.
Yost says that the new Nehalem machines will be priced at about the same price points as their predecessors in the IBM lineup using quad-core Harpertown chips. In some cases, because of the richness of the configurations, prices will even rise. (Well, at list price, anyway. We'll see about street prices).
The HS22 has been shipping to selected customers already, and will be generally available on March 31. The new Nehalem rack servers have also been going out to selected customers since March began, but will not be generally available until the end of April. The iDataplex nodes have seen some early testing customers, too, but will not be available until May. Specific pricing information will not be released on this gear until it ships. ®
COMMENTS
To the HP bigot
Actually I am not an IBMer, I am a customer - Major engineering company with over 1500 blades, all converted to IBM after a disasterous year with HP blades - over heating, drive failures, backplane failures (passive backplanes my butt !!) but what erked me the most was unethical sales practices by our HP rep. We had enough and turfed them out - you will soon see. If there's FUD out there, it's all HP. Like I said, it took HP 3 tries and 5 years to get to where IBM was 5 years ago. If you 've seen the NDA then you know you will have to rip and replace all your existing infrastructure in the next 18 months. Good luck
Re: Last page for response to Anonymous
Just to clarify the Itanium situation. Last readmap I saw had Itanium going for at least another 3 generations last IBM roadmap I saw had Power 6+ and didnt even mention 7 and that was 2 months ago! Oh and lets bring some real world economies into this, do you know how much it costs to start making a new processor design including new fab technology and R&D about 2 billion! Unfortunately Intel are VERY good at manufacturing processors and making it pay, to be brutally honest I dont think IBM can afford to stay in the game for much longer, particularly as their last balance sheet showed them making a loss in their microchip business!
Oh and INTEL dont care about MS products (well other than SQL) because they have Oracle, SAP etc on their side (just look at the licensing to see how unfair it is when trying to size for an IBM system) last one we did on our systems showed that just by moving from Power 6 to Itanium we would save 30% in licenses for our Oracle and at Oracles prices thats enough to pay for the hardware! As for performance dont go there, your just showing yourself to be the script kiddie you really are, as far as your concerned its all about MHz isnt it ...................... tit!
Anonymous - Get real and get your facts straight
Oh look an IBM salesgrunt! Either that or a user who is so far up IBM's a*se everything looks blue.
Well let me tell you mr IBM salesperson. I work for quite a large company who until recently was predominantly IBM. That is until we came to refresh time and started looking around to see what else there was.
I dont have time to address all of you "facts" but lets take a few and see how wrong you are.
"2) you want 16 blades for max density- sacrifice redundancy as each blade has only one connection for power and I/O to the midplane"
WRONG! You see what you fail to understand is that the midplane on a HP is passive which means there is nothing on it to go wrong, all of the components are in seperate modules easily accesible out the back. What happens if an IBM midplane has some of its components short out, oh sure there is redundancy but like when it happened to us you end up having to have the whole midplane replaced which means taking everything in the chassis down thats before you loose half of your I/O to every single blade in the chassis!
"6) Virtual Connect - proprietary, and 3 times the cost of IBM's Open Fabric manager"
You see this is the one that makes me think you are an IBM employee. Virtual Connect is an end point on the network, that means any networking device sees it as a NIC or HBA, the only thing proprietary about it is the HP badge on the front and the fact that IBM didnt think of it!
"7) Noise level - OUCH for HP, 68-72 decibels, IBM as low as 60."
Whats wrong you running a yoga class in your DC or something? Of all the complaints about kit this has to be the lamest ever!
"9) HP blades have thermal issues with memory dimms (run at 98degrees)
10) Oh and what idiot installs a hard drive on top of a heat sink (surely you gest !!!!)"
Ah the old power and thermals. Well all I can say is that our DC runs about 2 degrees cooler now that we chucked that BladeCentre rubbish out, oh and HP actually give us real time data on our power and thermal unlike IBM's guesstimate, which is quite useful when I have to talk to our facilities people!
"12) IBM did not announce the higest Nehalem speed bin and NEITHER DID HP"
Sorry just ordered it!
"13) Latest rumour that HP is coming out with yet another chassis design within the next 12 months once again adhereing to their RIP AND REPLACE strategy"
Ive seen the NDA, I would be very very worried if I was IBM!
Oh congratulations by the way on cutting and pasting IBM sales FUD, they really are masters of the brown stuff arn't they. If only they put as much effort into their products they wouldnt have to slag everyone elses off!
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
