Downtime is never a good thing for any business, except maybe for those who sell high availability clustering and disaster recovery software for a living. And to be honest, these companies explicitly want to keep the downtime and the recovery time to the smallest period as possible. The fear of downtime should be enough to motivate us all to replicate out systems.
But sometimes, just having two systems in a high availability cluster is not enough. Some companies need belts with their suspenders, and maybe even a bunch of different belts because they really can’t be down. To that end, Maxava has added a multi-node role swapping, or MNRS, feature to its eponymous Maxava HA line of clustering software. Specifically, this MNRS feature is available with the Maxava HA Enterprise+ edition of the software, which is aimed at very large IBM i shops that need more capabilities than the base Maxava SMB product delivers, such as MQ Series replication, QDLS replication, and simulated role swapping between source and target machines to actually see if the role swaps work. The base Mavaxa SMB product includes data replication between systems plus replication of over 50 different types of IBM i objects, spool file replication, Integrated File System (IFS) replication, and dynamic database management.
“The classic Maxava customer has one source machine and one target machine, and that target machine will tend to be a long way off, on a different power grid and sometimes in a different state or country,” explains Simon O’Sullivan, senior vice president at the HA and DR software maker. “More recently, that backup machine in a cluster can be in the cloud. Some of the larger IBM i customers, however, have several boxes – and the ones that we are dealing with are usually large banks and large insurance companies, and they have one production machine and two targets. One of our customers is in both markets, and they have production banking and insurance applications on different logical partitions on one machine but split the HA clustering to put banking on one target and insurance on another target. In other cases, it is quite common to have a disaster recovery machine and maybe even a test and development machine that is separate from the source production machine and the target replicated machine, and companies want to keep these systems all synchronized in some fashion.”
Hence the need for the MNRS feature. By the way, there is no architectural limit to how many machines can be in an MNRS broadcast set, according to O’Sullivan, and for those with even more complex hierarchies of HA and DR coverage, boxes can be daisy-chained off each other in complex ways, too. The MNRS feature keeps everything in the group absolutely synchronized, while daisy chaining would add some latencies as replications cascade outward from the production systems. The point of MNRS is that any of the multiple target machines in a group could be used to recover for the production machine in the event of the failure of that production machine, and the role swapping happens just as automagically as it would if there was only two machines. Obviously, you would not recover to both machines at the same time; one would be designated the new temporary production machine and the other would remain a target in the HA cluster.
In addition to the MNRS feature, Maxava has also boosted the performance of the replication of the IFS file system so it takes advantage of the multithreaded nature of modern Power processors like the Power8 and Power9 chips. By exploiting the multithreading, the IFS replication feature in the Enterprise+ distribution can dynamically replicate up to 255 processes in parallel (that is the upper concurrent thread limit of a logical partition in IBM i, so that is not a limit that Mavaxa has, but rather that IBM i itself has). We are not sure how much of a performance boost this parallel IFS replication gives, but we assume it is pretty substantial provided that the links between the two servers are not starved for bandwidth.
Unlike many systems programs on the IBM i platform, Maxava charges for its software on the IBM i and OS/400 software tiers that everyone is familiar with, and does not count logical partitions or cores like many licenses for systems programs and applications do. This makes the Maxava pricing as neutral as can be to adding cores or sockets. Most of the company’s customers are in the P05 through P20 tiers, and getting Mavaxa HA Enterprise+ for two machines, including a year of support, will cost somewhere around $25,000. Depending on some of the possible add-on features, getting the same setup for two P10 class machines runs somewhere from $35,000 to $50,000, and systems in the P20 tier will run about $100,000 with a year maintenance for a pair of boxes. This may sound like a lot of money, but that price has come way down since Maxava entered the market two decades ago – and largely because Maxava entered the market, by the way.
Incidentally, Maxava software licenses don’t have hardware upgrade fees built in, unlike some systems and application software sold in the IBM i market. If you stay in the same IBM i and OS/400 software tier and move to new iron, as long as you are on maintenance, you just move the software over and turn the old machine off.
This article is written by Timothy Prickett Morgan, President of Guild Companies Inc and Editor in Chief of The Four Hundred. #tim4maxava
Reprinted with permission from Guild Companies. Copyright 1996-2019. Please subscribe to IT Jungle’s free e-newsletters at https://www.itjungle.com/subscribe.
Like us or follow us on