Both machines are meeting sales targets in the first year, according to IBM sources.
As the name suggests, the iSeries for HA machines are full-blown servers designed to be used as auxiliary hot-swap machines for customers that run one of the high availability software packages from DataMirror, iTera, Lakeview Technology, Maximum Availability, Trader’s, or Vision Solutions.
For years, IBM and its high availability partners have been giving out special rebates on hardware for companies that buy HA software and a new OS/400 server to support it, and the iSeries for HA box more or less made these deals a permanent thing by turning a promotion into a subset of the product line. The permanence of the iSeries for HA boxes is an important thing for customers. The rebates that IBM has traditionally given mean that customers have to first lay out cash and then wait to get it back to acquire a discounted box aimed at supporting an HA cluster.
The iSeries for HA products are designed to protect production OS/400 servers against outages related to planned downtime (during upgrades and data archiving, for instance, when systems have to be taken offline) and for unplanned outages (usually caused by software crashes or human error). IBM announced five different iSeries for HA machines: one Model 825, two Model 870s, and two Model 890s. In March, after receiving requests from customers who didn’t need such large servers, IBM expanded the iSeries for HA line to include a Model 810 server.
While IBM launched the new Squadron Power5-based eServer i5 machines in May, it did not roll out eServer i5 for HA editions until the end of July. Specifically, IBM has two versions of the Model 520 and four versions of the Model 570 that have been given the HA designation, along with the related price breaks, which cut roughly one third off the list price of the iSeries or i5 box. Customers negotiate the price down from there, of course.
The important thing about the iSeries and i5 for HA servers is that they are real OS/400 servers, and they are absolutely intended to be used as a rollover machine in the event of a system outage. They run OS/400 Enterprise Edition and have the ability to support 5250 workloads for the number of processors activated in the boxes. For the i5 versions of the machines, the servers support AIX and Linux in logical partitions, the full Virtualization Engine virtualization features (including micropartitoning), and have all of the capabilities of a regular OS/400 server.
They even have upgrade paths into regular i5 servers: first you pay to convert the i5 for HA machine to an i5 machine running i5/OS V5R3 Enterprise Edition (presumably the cost of such an upgrade is the difference in price of the machines) and then you can upgrade further up the i5 line using normal upgrade paths.
A variant on this theme, geared more for disaster recovery than high availability, is the iSeries Capacity BackUp, or iSeries CBU. Rather than being a machine that is intended to be used in a rollover situation for high availability, the iSeries CBU is intended to be used as a remote machine in the event that a disaster (either natural or man-made) wipes out your data center or otherwise knocks out your OS/400 servers.
The iSeries CBUs, announced a year ago with the iSeries for HA boxes, have a minimal amount of processing capacity turned on – just enough to run the replication software necessary to keep applications and data in lockstep with the production machine. In July, IBM announced a single i5 Model 570 CBU machine. It is unclear whether the company plans to launch smaller ones based on the two-way Model 520s and four-way Model 550s, or larger ones based on the Model 595s.
The iSeries and i5 for CBU machines have much lower price tags than regular iSeries and i5 servers, and their prices are significantly lower than even the iSeries and i5 for HA boxes. That’s because they have a lot less capacity activated than the iSeries for HA boxes. In the event of a disaster, you activate that remote iSeries or i5 machine and start running your workloads on it.
During a disaster – the river floods and takes away your data center – you can activate all of the processing capacity in the CBU box to support your workloads. However, adding main memory and storage is not free during this disaster (even though they can be added on demand), and after a disaster is over, you have to pay use fees for the activated processors in the box. By IBM’s definition, a disaster is something bad that lasts for more than four hours and can only be recovered from at the remote location; a bad memory card or a software crash does not qualify. With the Model 570 CBU, you have 42 processor days to test your disaster recovery procedures free of charge, and IBM recommends that you do.
These HA and CBU servers have been on the market for a little more than a year, so how are they doing? Steve Finnes, enterprise server and technologies segment manager for the iSeries Division, and the person who steers the high availability initiatives in the iSeries market, says that he had a certain target in mind when the boxes were launched and that the machines have hit that target. While Finnes was not at liberty to give exact box counts for the machines, he did offer some insight into who is using them and why.
According to Finnes, there are about 4,000 OS/400 shops worldwide engaged in real high availability (meaning they are looking for 24/7 availability for their production systems and have paid for secondary systems and HA software from one of the seven vendors). While IBM has never been specific about how many OS/400 server shipments HA software drives every year, it is obviously a major component of the marketing effort, in terms of the revenue it drives, even if the box count is relatively low and the share of the customer base (by my math, it’s about 2% of the OS/400 base) is fairly small.
Not everyone who gets HA today will buy an iSeries or i5 for HA box. Sometimes, rather than upgrade their main production server, customers keep their existing box and use it as a target machine in an HA cluster, and then buy a whole new box. So, for instance, a customer might keep its Model 870 as the hot spare and buy a new Model 570. However, doing so means upgrading the operating system on the older box (which is a hassle) or not upgrading in order to keep it in lockstep with the production machine (which is also a hassle).
Finnes says that customers are coming to the conclusion that it is often easier to just buy two boxes that are the same, one regular version and one HA version that matches it. In fact, he says that of all the HA-related iSeries and i5 sales IBM has done this year, the iSeries and i5 for HA boxes are accounting for anywhere between 15% and 20% of iSeries hardware sales. Considering that the product is fairly new and is offered at a discount, compared with regular iSeries and i5 servers, this is not bad.
While not committing to any future targets (at least not publicly), Finnes says that the rate of shipment for the HA variants is accelerating on an exponential curve in recent months, matching the ramp of the i5 servers themselves. Given this, he is hopeful that IBM can sell a lot more i5 for HA boxes. Finnes says that IBM sold as many i5 for HA boxes in October as it sold in the prior three months.
The iSeries and i5 CBU servers, while having a lower price tag, seem to be a harder sell at the moment. Finnes says that the iSeries or i5 for HA servers outsell the CBU servers by a ratio of 10 to one. If you are really serious about high availability, then you are serious about role swapping between servers, and that means the CBU is not a good fit. Most companies are more worried about the planned and unplanned outages and how to maintain 24/7 operations; they want this first, it seems, and then they will worry about the disasters and the resulting downtime they can’t control.
Because running HA software across a data center and a very remote location (which could survive a local disaster because it is nowhere near it) is problematic for the transactional workloads that OS/400s typify, it is not a simple matter to try to accomplish both high availability and disaster recovery with two machines. More things will go wrong within a data center, which is usually filled with cranky servers, and sometimes cranky people, and do so more often than the disasters that will completely take out a data center. The sales ratio of HA and CBU boxes reflects this, and it makes perfect sense.
Having said that, if a CBU box cost only a fraction of the cost of a whole iSeries or i5 server, and was available on a subscription basis, and dedicated to specific customers in a hosted model, I think IBM might make quite a bit more money than it is doing now from the CBU offering. No one wants to own a CBU box. They want to rent it, and they don’t want to pay a lot for it, since the odds are very small that they will ever actually use it.
