We Need A Baby PureSystems Running IBM i

December 10, 2012 Timothy Prickett Morgan

It is the holiday season, the time of giving. And, not coincidentally, the time of asking. I figured out a long time ago that you never get what you want if you don’t ask for it, although I must admit that at least in my life, asking has certainly been no guarantee of receiving. But asking is the first step. So here goes. What I want in 2013, aside from world peace, a pony, a little red wagon, and a sailboat, is an entry PureSystems machine that is actually suitable as the main data processing engine for the vast majority of IBM i shops.

This is not the first time that I have mentioned this. As soon as the Flex System chassis was launched in April, it was immediately obvious that for many small and even midrange businesses the Flex System chassis, which can cram 14 single-width compute nodes in a 10U space, was overkill. As was the whole idea of a rack-based machine. As the AS/400 and its progeny demonstrate all too well, SMBs like tower systems where everything plugs into the tower and you can tuck the machine into the corner of the office, a closet, off to the side on the manufacturing floor, or out in the warehouse.

Don’t get me wrong. I like the modular nature of the Flex System hardware, which jams server, storage, and networking into the same box, all under the thumb of a single management control freak and all integrating through a system midplane. It is what would happen if you took distributed systems and tried to turn them into an AS/400. But the fact remains that the Flex System iron and the PureSystems “infrastructure clouds” that are based on it are too much for your typical Power 520 or 720 shop with a few X86 servers running Windows. Even moving to the Storwize V7000 disk arrays, which were just pulled inside of the Flex System chassis as part of the November 13 announcements, is not necessarily the right storage choice for IBM i shops.

The word on the street is that IBM is, in fact, working on that latter issue. With the November switch and storage announcements, IBM also rolled out a sidecar disk expansion unit called the Flex System Storage Expansion Node for the System x240 Xeon E5-based server nodes. This is a complement to the PCI Expansion Node that was announced back in the summer. The Flex p260 Power7 and Power7+ nodes and the Flex x220 and x240 nodes all have a port–and I swear I am not making this up–called the Everything-to-Everywhere connector, or ETE in the Flex lingo, that allows for the PCI-Express bus to be extended off the compute node and out into the sidecar. The SEN and PEN units hang off this.

Here’s what the Storage Expansion node looks like:

At the moment, the Storage Expansion Node only links to the Flex x220 and x240 processor nodes, but the rumors I am hearing are that this storage sidecar will be supported on Flex p260 nodes running IBM i. That Storage Expansion Node would be useful for IBM i shops a couple of reasons. First, the disks in the unit are hot-swappable, unlike the two drives tucked under the lid of the p260 nodes. (The x240s have two hot-plug drives in the front, like most blades, but there isn’t enough room for them with the Power7 and Power7+ server modules because the processor sockets and heat sinks are too big.) Moreover, a single expansion module can hold up to a dozen 2.5-inch SAS or SATA disk drives or solid state disks, plus a variety of ServerRAID M5100 disk controllers with different degrees of RAID data protection. It is also fairly inexpensive (by IBM’s standards) at $1,669 not including disks or RAID controllers.

If all you need is a box of disks rather than a V7000 storage array, this is probably a better option.

The V7000, because it has all this funky thin provisioning, snapshotting, data compression, and replication software (among other things), is not cheap, at $14,500 for the Flex V7000 node with a single controller. It holds two dozen drives, which range in price from $459 to $1,109 each, depending on capacity, speed, and type. And the SSDs are crazy stupid expensive at $4,869 for a 200 GB unit and $9,369 for a 400 GB unit. Say you want a double-wide V7000 array inside your Flex System and you want to use 300 GB 15K RPM 2.5-inch drives. You’re in for $20,616 for the drives alone (if you load up the two dozen) plus $14,500 for the disk node and controller, for a total of $35,116. That’s 7.2 TB of capacity at a cost of $4.88 per GB, and that is even before you put the V7000 software on the box. The base Storwize file system is $11,000, and all of the add-ons would add another $14,000, for a grand total of $60,116, or $8.35 per GB.

I am not saying that such a Storwize setup is not worth the money or competitive. But I know my IBM i shops. They pay a premium for the operating system, database, and add-on data replication tools from third parties and they just want cheap internal storage that runs fast. And small shops probably don’t need 67.2 TB of disk capacity, either, not for a four-core server node. So a Storage Expansion Node that is driven by IBM i (very likely with the Virtual I/O Server embedded in the PowerVM hypervisor sitting in the middle) is probably a better idea for such shops. This would cost $11,977 for a dozen drives and the SEN, or $3.33 per GB. Granted, that’s still not cheap, but it is better suited to the kind of workloads.

While this would be good for IBM i shops with modest Power workloads and lots and lots of X86 workloads, the current Flex System chassis is too big, even in the SEN is attached. So I am proposing a different kind of machine. Basically, I want to take my katana, hold it high above my head, and cut a Flex System chassis with V7000 storage nodes in half, putting in p260 nodes, x220 nodes, Storage Expansion Nodes, and half V7000 nodes (call them the V3500) in the box. It would look a bit like this mock up I made just for fun.

With this machine, you have a box that is still 10U high, which is around 17.5 inches plus the height for the tower enclosure and maybe wheels if you want it to be movable. There is even enough room to extend the tower enclosure up a few inches and leave room for an Ultrium tape drive and still get it to fit under a desk. With this hypothetical Flex System Tower, which I show with two x240 nodes, one p260 node, and two V3500 arrays, you could have server and storage in the front and enough networking and power supplies in the back to put a complete baby data center under your desk. (Think of it as an AS/500 if that makes you happy.) I don’t like the idea that the Flex System Manager requires a whole node of its own, since that eats up precious space and compute capacity, so on this machine, I am going to put Flex System Manager inside of a virtual machine and let you use the rest of the capacity to run Windows or Linux workloads.

By doing this, you could have one Flex p260 node and one SEN right underneath it in the chassis, plus three Flex x240 nodes and one V3500 serving those nodes right underneath it. This box would be a good balance for many Power 520 and Power 720 shops–and importantly, would bring all the workloads under one management framework.

I thought that, more than anything else, was the whole point of “Project Troy.” Well that and making some money and fighting back against Cisco Systems. Let me give you some advice, IBM. Launch a Flex System Tower before Cisco launches a Unified Computing System for SMBs. Go after your own customers before Cisco does.