IBM Bolsters Disaster Recovery With GDR For IBM i
May 3, 2017 Alex Woodie
IBM next month plans to start selling a new disaster recovery product to IBM i shops. Called Geographically Dispersed Resiliency, or GDR, the new offering is designed to give companies an easy and affordable way to recover production IBM i LPARs on remote machines. IBM’s resiliency guru Steve Finnes gives IT Jungle the lowdown on the new tech.
GDR provides DR protection by essentially taking a production logical partition (LPAR), also called a virtual machine (VM), from one IBM i server and restarting it on another IBM i server. The VMs involved in a GDR switch must be stored on a storage area network (SAN) server – either an IBM DS8000, SVC, or Storwize server or an EMC array – and the two arrays can be geographically dispersed (as you may have guessed from the name).
Finnes says GDR serves a need for an affordable and easy-to-use DR solution. “It’s very, very simple and yet it gets the job done,” Finnes says. “One person can literally run a DR restart operation with this technology.”
Operators work with GDR from their PowerVC consoles, where they issue commands to an AIX LPAR called K-SYS. “The operator issues a command to K-SYS, which stands for control system, that says ‘I want to restart at site two.’ K-SYS takes over and conducts an orderly shutdown of the source and brings it up on the target side,” Finnes says. “The whole process is very simple.”
The software, which has been available for AIX and Linux customers since last year, is similar in some respects to VMware‘s popular Site Recovery Manager (SRM) product, which many X86 shops use to recover production environments in the event of actual disasters and to conduct DR readiness checks.
Testing DR processes has been a big point of emphasis in the business resiliency field, and so DR readiness is a big part of GDR, according to Finnes. “Typically what people are doing in the DR site is testing. They need to test to demonstrate they can do a DR operation,” he says. “Hopefully nobody ever actually has to use the disaster recovery. We wouldn’t wish that on anybody. But the ultimate purpose of any DR solution is if I did have a disaster, I can actually go and recover.”
The GDR product, which becomes available for IBM i on June 23, borrows concepts and technology from other IBM offerings that many IBM i shops are familiar with, including Live Partition Mobility (LPM) and PowerHA.
GDR is “conceptually a first cousin [to LPM]. The basis is essentially the same,” Finnes says. The big difference, he says, is that there is no rebooting the VM with LPM.
“With GDR, I’m essentially going through an IPL process,” Finnes says. “I’m actually creating those VMs on the fly, bringing it all up into the processor complex, loading up the environment, and resuming production on that target box.”
While GDR is separate and distinct from PowerHA, it does borrow data replication technologies from PowerHA. Specifically, it uses the Metro Mirror and Global Mirror replication protocols to actually transmit data from the storage array in the production environment to the storage array in the backup environment. For EMC environments, GDR uses the Symmetrix Remote Data Facility (SRDF) replication protocol.
A GDR customer would use the synchronous Metro Mirror protocol when the production and backup systems are located on the same campus or city, and use the asynchronous Global Mirror protocol when the sites are distantly located. SRDF supports both synchronous and asynchronous modes.
In some respects, recovering an IBM i environment with GDR is similar to recovering an IBM i environment with tape. “That’s essentially what it is. You’re really rebooting or IPL’ing that VM into the target site,” Finnes says. “You do a scratch install, boot up everything off of this image.”
Because of the need in GDR to wind down the VM on the production box in a methodical manner and then spin it up in the new environment, the recovery takes a bit of time. “It’s going to take longer [than a high availability role swap] because you’re going through essentially an abnormal IPL process,” Finnes says, adding that the time it takes to complete the IPL will vary according to the database size and other factors.
In terms of recovery point objective (RPO), the GDR solution will be very aggressive in recovering all of your work, or nearly all of it. If GDR is configured with Metro Mirror or synchronous SRDF data replication, the RPO could be as low as zero, meaning that as soon as a transaction is written to the local machine (or the local storage server, since that’s where the VM actually lives), then it’s protected with GDR.
The recovery time objective (RTO) is not nearly as aggressive with GDR, which is what you would expect with a DR solution. Because a GDR recovery is essentially an abnormal IPL of the production IBM i system, it could take some time for the customer to get back up and running with GDR, as Finnes previously mentioned.
These characteristics make GDR a good potential solution for customers who are looking for a DR solution. “There are people who want a DR solution. I think it’s going to be interesting for those people because it’s relatively inexpensive and it’s very simple to use,” Finnes says.
The one caveat to GDR is that it does require the user to be invested in IBM or EMC storage arrays, and be somewhat capable in AIX technologies like K-SYS and the virtual I/O server (VIOS). Because many of the IBM i shops that have bigger IBM SANs, like the DS8000, they are likely to be using PowerHA – and because PowerHA already includes comprehensive high availability and DR capabilities – that limits the potential market for GDR among those bigger IBM i shops.
“The DS8000 customers are really, truly mission critical, the ‘who’s who’ of American industry,” Finnes says. “There’s no reason for them to do this. They already have a compressive HA and DR setup within the cluster so I don’t envision the DS8000 clients doing this.”
On the other hand, smaller IBM i shops running the popular Storwize arrays are not as likely to be invested in PowerHA, and therefore could be good candidates for GDR. As of June 23, the SVC storage server – which is not nearly as popular among IBM i shops – is now supported with GDR, along with IBM i, Storwize, and DS8000 arrays.
There is also the potential for IBM business partners to build DR as a service (DRaaS) offerings atop an IBM storage array with GDR supplying the replication and recovery capabilities. Finnes says he’s seen a lot of DRaaS offerings pop up recently, which makes him think there could be a good opportunity here for GDR.
GDR potentially could reduce the software licensing costs for customers compared to PowerHA or high availability solutions built atop logical replication technology. Because the customer’s application software is never active on multiple machines at the same time, and the secondary system is a true “cold” backup – as opposed to a “warm” or a “hot” backup, which is often the case with HA or clustering technology – it could soften the software licensing requirements for customers.
The list price for GDR is $1,020 per core for a “small processor” and $1,575 per core for a “medium processor.” Please contact IBM for definitions on processor size.