Mainframe Storage

Nick Clayton posted Mon February 03, 2025 04:53 AM

The most important difference for the DS8000 G10 system is the change in the RAID adapter and drives. So you need to consider which drive type to use here and in almost all situations you want to use a single type and size of drive. The main choice is whether you want to take advantage of FlashCore Modules and compression or to use Industry Standard drives. 

If you are replacing an existing system then in general you should consider larger drives than the current system as the new NVMe drives provide higher performance per drive set. In general I would start with the same number of Host Adapters and equal or larger cache as the old system unless there is significant growth planned.

You might want to discuss the configuration and opportunity with your IBM Partner Technical Specialist to discuss the specifics.

JONGCHUL LEE posted Tue February 04, 2025 01:23 AM

If you have introduced FCM to 8a10 storage, please consider that encrypted data among the operational data will not be compressed and there will be no capacity reduction.

Ian Wright IBM Champion posted Tue February 04, 2025 09:37 AM

I have a blog up on the Mainline website

But I'll break it down somewhat

• For compression, make sure that you do sizing -- There's no "compression guarantee" for the FCMs like there is for a FlashSystem. My assumption is that this is because of the fact that any shortage is not going to be solved by adding a few drives, but would be an entire 16-drive install group. But sizing isn't necessarily straightforward.
  • For Mainframe, remember that most mainframe environments are doing at least some compression in the mainframe itself for both performance and capacity benefits. So, for them you're going to need to use the APAR for SMS. This is relatively new and I've had a customer tell me that he didn't want to install it yet as a result. But, without that information you really don't know how much compression you're going to get vs. stuff that's already compressed.
    The mainframe also has some environments making use of Pervasive Encryption. Encrypted data, like compressed data, isn't going to give you anything.
  • For IBM i, there's no compression tool. However, users can use the SAVLIB tool on a representative database and get a good idea of compressibility. Make sure that you don't count IFS data in your compression calculations since image data isn't going to do anything with compression.
  • Otherwise use some other compression tool (gzip, pkzip, winzip, etc...) or the Compresstimator.
  • Some people may just want to get FCMs and not do any estimation, expecting whatever compression they get to be a bonus. And that's fine, but realize that the FCMs are more expensive than industry standard drives. They do have more capabilities coming like Ransomware detection, but they need to be aware of this.
• Realize that the focus for performance/bandwidth improvement isn't on the front end of the system, but is on the backend with the new PCIe 4 based HPFEs. They have vastly more bandwidth but you might not see as much performance improvement on the front end when you're doing modeling in StorM.
• Along with that, what I found in my performance modeling is that I didn't see any performance difference between having a small number of 19.2 TB FCMs in a single HPFE pair vs having a system spread out into a second frame using smaller FCMs.
  • YMMV, but this modeling was done for two customers, an IBM i shop (currently going through implementation of their G10s) and a Mainframe shop that we're still working with.
• Keep in mind that it is no longer supported to have different tiers of Flash drives in a pool for Easy Tier and you can't have FCMs and Industry Standard in a single pool.
  • StorM started recommending against that a while ago, but now it's a rule. 
• The only "standard" parts for FCMs are currently 19.2 TB FCMs. Because there are (16) drives in a set, this is going to be quite a bit of capacity even before compression.
  • You can get 9.6 TB or 4.8 TB FCMs, but you need to have an iRPQ filled out for it (for those of us who are partners, IBM has to fill this out)
  • Since we can't have a pool split between different types of drives, you need to commit to the drive size that you're going to use from the beginning. If you might need to add capacity but it's going to be a relatively small amount, you may want to get the RPQ so that you can go with a smaller size. Smaller sized drives will also allow you to be more precise on the amount of capacity needed sometimes.

That's what I have off the top of my head.