Sundry Power Systems and BladeCenter Announcements
October 11, 2010 Timothy Prickett Morgan
The timing was probably coincidental, or the result of some vestigial impulse to make announcements during the fall COMMON midrange user group meeting, but IBM late last week announced a slew of new high-end hardware disk arrays (which don’t have much to do with most IBM i shops) and tweaked its Power Systems and BladeCenter lineup a bit.
IBM scattered these announcements all over the place last week, so let’s go through them one letter at a time. In announcement letter 110-199, IBM made some changes to both Power Systems and BladeCenter platforms.
On the Power 710 and 730 servers, IBM is reducing the required minimum main memory from 8 GB to 4 GB, and is also introducing a new memory feature, which is known as feature 4525, that is two 2 GB memory sticks running at 1.07 GHz. This memory feature costs $532, or $133 per GB and will be available on December 10.
The Power 740, which was previously only available in a base configuration with two eight-core, 3.55 GHz Power7 feature cards, is now available with a single feature configuration. It looks like IBM’s yields have improved on these eight-core chips at the same time customers were complaining about having to buy two feature 8355 processor cards. As I showed in last week’s issue, at 97,700 aggregate CPWs of online transaction processing power running IBM i workloads, a 16-core Power 740 was over the top anyway. IBM had already tested a Power 740 machine with only one card, and it is rated at 47,800 with all eight cores dedicated to running IBM i. For most shops, it is still less expensive to buy the processor feature and core activations a Power 710, 720, or 730 to get eight cores, and in many cases, as I show in this week’s issue, the software costs start eating you alive on the Power 730 and 740 configurations. That said, if you are worried about your ability to expand in the future, the Power 740 is basically your only option unless you want to make the much bigger jump to Power 750 machines. Anyway, this one-board Power 740 will be available on December 10 as well.
The Power 720 and 740 servers have a new PCI-Express 2.0 riser card and related Ethernet adapter card, too. Feature 5685 is a Gen2 riser card that has four PCI-Express low-profile peripheral slots. The low-profile two-port feature 5284 10 Gigabit Ethernet adapter card requires this new feature 5685 riser; this 10 GE adapter, which has SFP+ fiber cables, will not plug into the existing feature 5610 riser card. There is an alternative feature 5284 two-port 10 GE adapter that has copper twinax SFP+ cables, which also requires the new PCI-Express 2.0 riser card. The feature 5685 low-profile riser card costs a stunning $2,000, while the fiber 10 GE adapter costs $3,500 and the copper 10 GE adapter costs $1,800.
Not that this matters to most IBM i shops, but there are some telecommunications companies around the world who have to meet the stringent NEBS ruggedized, DC-powered system requirements, and IBM has also announced a new 1,700 watt power supply for the Power 750 (feature 7708) that makes the machine NEBS compliant. This power supply costs $1,500. IBM also said in a statement of direction that the BladeCenter PS700, PS701, and PS702 blade servers will be available in NEBS compliant configurations in the first quarter of 2011. IBM has to rejigger some of the thermal settings in the power management features on the Power7 blades to make them compliant with the NEBS specs.
IBM has also tweaked the SAS RAID controller with 380 MB of cache to create a new feature 5805, which has thinner backup batteries than the feature 5903 RAID controller that was announced earlier this year. (You can’t hammer these things into the chassis, after all.) These controllers are supported in the Power 5XX and Power 7XX lineups as well as in the Smart Cube appliances, which have gone the way of all flesh as far as I know.
IBM said also that on November 26 it will put out some PTF patches for IBM i 6.1 and 6.1.1 that allows its TS3100, TS3200, TS3310, and TS3500 LTO-5 tape libraries to be supported on these operating systems. IBM had been supporting the TS2250 and TS2350 LTO-5s through a bridge box already. These two earlier operating systems are getting support for IBM’s feature 4807, 4808, and 4809 PCI-Express cryptographic accelerators starting on November 19. These encryption/decryption engines were previously on supported on AIX 5.3, 6.1, and 7.1 and IBM i 7.1.
On the blade front, the BladeCenter H chassis now has a 2,980 watt power supply (feature 9058), which are sold in pairs and then doubled up to offer redundancy. The first two are part of the chassis, and the redundant pair cost $549.
The PS700, PS701, and PS702 blades as well as the older JS23 and JS43 blades using Power6+ chips now have a new two-port Gigabit Ethernet expansion card, feature 8243, which costs $299.
IBM is finally shipping the Nexus 4001I switch from Cisco Systems, which I told you about a year ago and which implements the Fibre Channel over Ethernet (FCoE) standard and uses Cisco’s IOS operating system and management tools, but inside of IBM’s BladeCenter instead of inside Cisco’s own “California” Unified Computing System blades and racks. This switch is not cheap, but then again, no switches are. The switch has 14 10-Gigabit Ethernet ports that internally link to each of the blade servers over the BladeCenter backplane, plus one 100 Mb/sec internal port and one Gigabit Ethernet external port for switch management, and six 10 GE external ports to reach out to storage and other server networks. The Cisco 4001I switch is known as feature 2241 in the IBM catalog and it costs $11,999. The software to add FCoE capability (what, you thought it would come by default and for free?) is feature 2242 and it costs an additional $3,899.
At these prices, you can see now what IBM has figured out it needs to be in the switching racket and has therefore acquired blade and rack networking partner Blade Network Technologies for a rumored $400 million two weeks ago.
In this announcement, we also learn that IBM is peddling Red Hat Enterprise Linux 5 pre-installed on various Power-based machines, including the BladeCenter JS12, JS23, JS43, PS700, PS701, and PS702 blade servers and the Power Systems 520, 550, 560, 570, 710, 720, 730, 740, 750, 755, 770, and 780 rack or tower servers.
Now, if we jump over to announcement letter 110-188, we see that IBM is peddling a two-port 10 Gigabit Ethernet converged network adapter card, feature 5437, plus Brocade’s 10 Gigabit Ethernet switch module for the BladeCenter H chassis. The Brocade switch has a total of 16 ports in its base configuration–14 out to all the blades in the chassis, plus two external ports. With a golden screwdriver software upgrade, you can turn on all 30 ports in the switch, which gives you eight 10 Gigabit Ethernet external ports and eight 8 Gb/sec Fibre Channel external ports to reach out to storage area networks. IBM has integrated its Systems Director server management tools to work with the Data Center Manager fabric management software from Brocade. Pricing for these two Brocade products were not available at press time.
As you can see in announcement letter 910-179, IBM is mothballing a bunch of expansion cards, switch modules, and pass-thru modules for the BladeCenter chassis and related blade servers effective October 31; there are other products in the catalog to replace them, but often not from the same vendors.
That leaves the enhancements to the PowerVM server virtualization hypervisor for Power-based systems. In announcement letter 210-196, you’ll see that IBM has tweaked PowerVM so you can do a suspend and resume of a logical partition for longer than five to 10 seconds and store the state of that partition (including its memory, non-volatile RAM, and virtual service processor (VSP) state) on disk or flash persistent storage. By freeze-drying a running partition, you can free up the memory resources dedicated to it and also bring it back to life on the same physical server or on another server after live migration (if you happy to be lucky enough to be using AIX). The suspend resume functions work for AIX and Linux partitions managed by a Hardware Management Console, but do not–surprise, surprise–work with IBM i.
(I am not a violent person by nature, but sometimes, I want to borrow my Dad’s size 13 cowboy boots and just start kicking asses in Somers, Armonk, and Rochester until I can’t move my legs any more. And then, I’d rest up for an hour or two, sipping a good strong stout, and then start up all over again just to make sure it all sunk in. You know what I mean?)
Virtual I/O Server 2.2, which is updated with the new PowerVM release announced last week, allows for the creation of shared storage pools for VIOS partitions that span multiple Power Systems machines. IBM has also tweaked VIOS to allow for so-called thin provisioning, allowing for resources to be added and subtracted from running VMs instead of being locked down statically when LPARs are created. VIOS partitions can also be grouped together to share a common storage pool and be managed as a single unit, which simplifies management of VIOS across multiple Power Systems machines.