After an upgrade, it's important to reflect on the change and compare your new environment with the old.
SSDs and HMCs are my new must-have.
We migrated from a pair of Power Systems 720s to new Power Systems S814s in January. For this upgrade, we altered our environment a few different ways from how we traditionally set it up.
The biggest change that affected mindset was the addition of a Hardware Management Console (HMC). For years, we relied on either a twinaxial console or a LAN console. I've wanted the HMC for a long time, but our partitioning requirements were really more "wants" than needs. We dabbled in Linux guest partitions but never had much real love for partitioning because we had to manage the partitions via System Service Tools (SST), which doesn't allow much flexibility. I viewed the HMC as more of a pivot point where we could really leverage our Power Systems hardware. Since getting the HMC, we've already spun up a new IBM i guest partition for the purposes of upgrading some WebSphere-based workloads by way of a side-by-side method, rather than upgrading them in place. This allows us to migrate the data to the new partition, test it, and then make sure everything is good to go before simply pointing DNS records to the new server. This type of compartmentalization has really opened the doors to the real power of partitioning, and the HMC makes it much easier to get a handle on.
So where's the affect on mindset? Well, you really view the operating systems and the hardware as the two separate entities they are when you work with the HMC. If you don't have the HMC, that separation is not very evident at all. You hear people talk about the separation of IBM i and Power Systems, but when you view the machine and the partitions running on them within the HMC console, things really come into focus.
I quipped on Twitter that I don't understand why IBM doesn't package up the HMC as part of the Power Systems configuration by default. I understand that cost is a factor, but when you look at the value you get with the HMC you really see that you're getting your money's worth. For instance, being able to remote power-on partitions is invaluable, considering I live about 45 minutes from my data center. You can't do that with a LAN console. Over the course of the three to five years you use a box, then it's a next-to-nothing cost...maybe a couple hundred bucks a month. It may be a good fit for you, or it may not. I suggest that you at least evaluate the benefits of a HMC before your next upgrade, especially the partitioning components. If you're looking at Linux partitions for additional workloads or IBM i hosting IBM i, then you really need to have a good look at an HMC.
Another big alteration in our environment is the switch from spinning drives in our system Auxiliary Storage Pool (ASP) to Solid State Drives (SSDs). Our independent ASP (IASP) is still all spinning drives, and if I had my time back I would've found a way to make the case to change those to SSDs as well. One of the main reasons is that if I put any of my guest partitions in my IASP for high availability with POWERHA, I'd want them to perform just as well as the partition that runs my system ASP. It's not like the IASP isn't fast, but it could be much faster. It reminds me of when Capacity Upgrade On-Demand came out and a customer of mine at the time joked that if he temporarily turned on an idle core, he wouldn't want to turn it off. I've seen the speed of SSD, so I want it on everything now. If only I had an unlimited budget.
Truth be told, SSDs are a bit more expensive, but I've never seen an IBM i partition fly like this before. It's smoking hot. Think the surface of Mercury hot:
- Our nightly backups on the 720 used to run about three hours. With the SSDs in place on the S814, they run one hour and fifteen minutes. This isn't virtual tape (i.e., saving to disk) either. This is straight to a six-year-old TS3100 LTO4 library.
- We had some long-running jobs that took about 30 minutes overnight that now complete in six minutes.
- Applications such as IBM Domino and IBM Connections, which reside entirely in my system ASP, demonstrate increased performance and ridiculously fast response times for web interfaces.
- Our IPL went from about twelve minutes to five.
Of course, you can always add one or two SSDs to your environment and then use a number of ways to get your hot data onto them. I went full SSD on that ASP for simplicity because most of the data on that ASP is hot. There's no management of moving data to consider whatsoever.
I'm sure the change from older POWER7 processors to POWER8 had an effect on performance, but I've never seen more of a change from system to system than moving to solid state from spinning drives. We were never really processor- or disk-constrained anyway. We were memory-constrained for the most part. So for the purposes of this exercise, we took a good disk-performing system and made it an unbelievable disk-performing system.
Another component of our installation you may find interesting is that one machine was built in and shipped from Mexico and the other from China. Both racks took a little longer to arrive than expected, but they were identical in terms of hardware placement, right down to the Velcro strap locations in the cable trays. The only minor difference we could find was with the keyboard touchpad on one of the HMCs. One HMC keyboard didn't have one, but it did have the eraserhead mouse, so we weren't halted in any way. I just need to call IBM support to have them bring one in. The nearly identical machines coming from two factories twelve thousand miles apart is indicative of great quality control. While I'd like to see these machines built in Rochester, Minnesota, the fact of the matter is that they're made elsewhere. Concerns about the quality of the hardware and the meticulous care of the build are so far non-existent with my machines. This is an important note for all customers.
The only real snag in our migration was from a POWERHA perspective. I won't get into it in this article, but look for a TechTip in the near future about what challenges we faced and what we did to perform a successful migration to POWER8 on two machines. We complicated matters by moving our disaster recovery site to another physical location, requiring IP address changes and a pretty good shaking up of our POWERHA setup during the migration.
I want to give a big thanks to iTech Solutions who were a great resource and a pleasure to work with in getting these systems to POWER8. They're just top notch.