Which Way to HA? Hardware, Software, or Both?
June 14, 2010 Dan Burger
The availability of data and applications is hyper-critical for many organizations. Therefore, the decisions that are made when architecting a HA strategy become hyper-critical as well. The goal of minimizing downtime to the skinniest possible margin is the challenge for the top global corporations, but the pathways to reaching recovery time objectives (RTO) and recovery point objectives (RPO) are being closely watched by all companies concerned with business process management. In the IBM Power Systems community, this is particularly interesting.
“The business requirements come down to RPO and RTO,” says Ron Peterson, senior product strategist at Vision Solutions. Peterson is the author of a recently published white paper that examines two methodologies for achieving RPO and RTO goals. One method is based on hardware and is currently being championed by IBM. The other is based on software and is the favored technology of Vision Solutions.
Given the dichotomy of these two approaches and the sometimes incendiary nature of the high availability purveyors, you might expect this white paper to stick a knife in the back of hardware-based HA. Not exactly. Peterson explains that “both approaches are strategic for IBM i and Power Systems hardware, and both have advantages as well as challenges that must be overcome in order to meet the goals of no loss of data (RPO) and the fastest possible recovery (RTO).”
Among the explanations and distinctions being made in the white paper is that a hardware-based system consists of either page-replication at the operating system level or disk-sector replication at the storage area network (SAN) level. For companies that have already moved into SANs and are sharing external storage with other IT environments within their organizations, hardware-based HA fits well with a plan that is external storage oriented. (That’s my observation. It’s not written into the white paper.) And that’s not to say software-based HA won’t work in a SAN environment. It most definitely will. But a SAN system is hardware focused, so hardware-based HA seems to be more in line with the thinking of SAN-minded project managers who are enthusiastic about related technologies such as independent Auxiliary Storage Pools (iASPs).
It’s not coincidental that IBM has endeavored to boost external storage performance since the introduction of Power6 servers and its emphasis on storage area networks has moved along with that. IBM’s acquisition of DataMirror, a high availability and disaster recovery vendor with years of AS/400 experience in 2007 plays into this as well, providing IBM with software-based HA technology.
Peterson says companies that are moving toward external storage should not lose sight of the benefits software-based HA bring to that arena. Bandwidth and latency are better controlled using the remote journaling capabilities that underpin Vision’s software. And recovery will be quicker and easier, Peterson claims.
Because logical replication allows a selective process, a substantial amount of replication is eliminated because there is no need to replicate every bit of data or every file. The extent of the selection process cannot be matched with hardware-based HA, Peterson says, and therefore the software hogs less bandwidth. Not only can software-based HA minimize the amount of data being transferred, but it can optimize data transfers as well by bundling journal entries and streaming large block sizes when journal entry transmission lags production activity–a latency issue.
Logical replication runs above the operating system. According to this white paper, this is an advantage because it has priority status when accessing system resources. It monitors and reads journals (think of them as logs) and applies the changes defined in those journal entries to a second system, commonly referred to as the target system. The target system is sometimes a virtual system within the same machine but in a logical partition, sometimes a co-located physical system, and other times it’s a separate physical system or LPAR on a system in a separate data center.
As IBM has enhanced the i operating system, logical replication technology has realized better performance, expanded the reporting capabilities, and improved management capabilities.
The white paper also highlights software-based HA advantages such as real-time, read-access to the target-side data, a useful feature when querying data or performing offline saves to tape. Hardware-based systems have limited accessibility and, due to a larger amount of data being transferred (minimal selective processing), the source and target systems take longer to get in sync.
Additional red flags are raised in this report for those considering hardware-based HA. Among the cautions that Vision believes are important are: potential data synchronization problems, sharing SANs among multiple platforms, an iASP switching problem that affects pre-IBM i 6.1 versions of the operating system, and the inability to replicate a list of 17 specific SYSBAS objects.
One of the red flags that Peterson was waving when I spoke with him last week was that hardware-based HA was being perceived as the better choice for HA because it was the newer choice–newer being better in some people’s minds.
“The white paper was written to point out there are some challenges and there are some technology differences to take into consideration,” he says.
One of the conclusions I jump to when I hear Vision or other vendors express their concerns about IBM crowding their territory is that IBM can be very influential with its customers. If Big Blue chooses to push hardware-based HA, for instance, as the best-in-class solution, other options can get smothered. This isn’t just a debate between hardware and software choices or a spin-off argument about external or internal storage technologies. It seems to me that Vision is trying to prevent one-sided conversations on the topic of HA. Their people don’t want to see Vision’s product’s strengths ignored. IBM’s competition with its business partners tends to make these situations a fight with kid’s gloves . . . at least when it’s a public forum.
Six weeks ago, at the COMMON Fall Conference and Expo in Orlando, Florida, I talked with Bill Hammond, Vision’s director of product marketing, about the divide between hardware- and software-based HA. Hammond pointed out that it’s not an either/or proposition and that his personal opinion is that hybrid systems–a combination of both software and hardware–will become popular.
“The solutions that are going to be most prevalent in coming years will be hybrids of hardware- and software-based HA,” Hammond said. “It’s not a mutually exclusive conversation. If a company is committed to one side, it doesn’t exclude the other. You have to look at each location and each application separately and then make a decision on what is best.”
You can grab a copy of the Vision Solutions white paper, titled High Availability for IBM i: A Technical Review of Software- and Hardware-Based Solutions, here.