When apps attack, memory is the weakest link.
IBM announced the new 4-core POWER8 S814 on June 10. While the new P05 iron works well for many shops, I think a lot of customers could benefit from added memory capabilities if IBM were able to do it.
Back when IBM announced the first round of POWER8 boxes in April, a number of people were wondering where the true SMB box was. By SMB, I mean in the P05 software tier. Most of the IBM i on Power Systems market is around the P05 size, small machines with a couple of cores running the small side of the small-to-medium businesses. Those are the IBM i bread and butter customers right there.
Because I have a non-disclosure agreement (NDA) with IBM, I knew that box was coming and even had some good conversations with IBM about its features. At first glance, it looked like it would be the box I'd be able to jump down into from a P10 since I use only two cores. Yee haw! My only concern was the lack of memory. At 64 GB of RAM, it was just too small. I'd be looking at two boxes to do what I do now, and a third for growth. To be fair, the 4-core POWER7+ equivalent had a hard memory cap as well, so it shouldn't be much surprise to anyone that if they wanted to do memory-intensive applications, then they'd have to move to one of the larger 6- or 8-core offerings. The S814 supports only four 16 GB EM8B memory cards. Granted, it supports only eight disk drives, but with eight 387 GB SSDs, you're not too limited on disk space with close to 4 TB available if needed. I think the ratio for disk to memory needs to be changed a little to accommodate shops that want to add extra, non-storage-intensive workloads.
On those little boxes, memory is the biggest constraint.
This is a point of contention for me. I don't have too many users, but we run a lot of applications on our little 6-core Power 720. The nature of those applications (Domino and WebSphere) means we need a good chunk of memory.
Right now, I'm considering my upgrade options into POWER8 and weighing the pros and cons of adding external storage, specifically for sharing storage with some of our Windows workloads. We need to upgrade our Windows-only SAN anyway because the old one is dying a slow and painful death, with no disk caching at this point.
Then I started thinking, "Well, what are those workloads?" And can those workloads run on Power Systems eventually?
Active Directory, for instance. We have a couple of Active Directory servers that provide domain services, DNS, policy enforcement, and the like. So with IBM i 7.2, we've been gifted with Samba support. Granted, it's version 3.6, and the real benefits that I want are in version 4. As per the Samba 4 announcement, "Samba 4.0 comprises an LDAP directory server, Kerberos authentication server, a secure Dynamic DNS server, and implementations of all necessary remote procedure calls for Active Directory. Samba 4.0 provides everything needed to serve as an Active Directory Compatible Domain Controller for all versions of Microsoft Windows clients currently supported by Microsoft, including the recently released Windows 8."
Some other great features include Group Policies, Roaming Profiles, Windows Administration tools, and support for OpenChange, an open-source implementation of Microsoft Exchange Server, allowing complete interoperability with existing Exchange servers and Microsoft Outlook clients. Samba 4.0 also supports direct integration with Microsoft Exchange servers.
Not only can a Samba 4 server join an existing set of Active Directory domain controllers, but also existing domain controllers can join a Samba 4 Active Directory Compatible Server's domain.
The inability for us to move to another platform like Samba for a true domain controller was always due to the lack of Group Policy features. Samba 4 now allows Group Policy control of workstations running Microsoft Windows, which means we can potentially migrate a number of Windows servers to IBM i. Imagine that: IBM i as a full-fledged domain controller.
That's pie in the sky, and there are limitations with the current implementation of Samba on IBM i. For instance, there's no Kerberos support for Samba on IBM i, and NetServer won't start if Samba is running. But having Samba running natively on IBM i gives us a good indication that we'll be able to do domain controlling in the future. That means killing off unneeded Windows servers and bringing those workloads to IBM i.
Samba is also an alternative for file serving. While NetServer was a decent alternative a few years ago as it used Kerberos and it integrated with Windows fairly well, it seems that Windows 7 had more trouble with Netserver than I'd want to see, especially when EIM was involved. I've not been able to connect to Windows 7 computers with as much ease as I did with XP. Perhaps if IBM gets Kerberos working with Samba in the future, then the ability to replace NetServer with it will become a reality. Currently, it appears that it's for "basic" file and print sharing. Without Kerberos support, it just wouldn't be seamless.
IBM Notes Traveler was another big announcement about a year or so ago. It's a fantastic solution that takes away the need of having a Windows box running BlackBerry Enterprise Server (BES). I'm down to my last user on a legacy BlackBerry. In three weeks, that user will be outfitted with a new smartphone and moved over to Traveler, and we get to power down two Windows servers that used to run our BES workloads. IBM Notes Traveler isn't memory-heavy like say, WebSphere, but it does require allocated memory threads for each connected phone. It's certainly not storage-intensive. All Domino mail files are stored on their respective Domino mail servers. There's no mail replication to the Traveler server at all. The Traveler server acts as an intermediary between the mail server and the mobile device. It doesn't need disk. It needs RAM and processor.
Between file serving, domain controllers, and BlackBerry Enterprise Servers, that's six Windows servers that could be (well, the two BES servers actually are) taken over by IBM i in my environment. Which brings me back to my original argument: you need memory to run more workloads.
The value proposition for IBM i is the ability to consolidate and integrate. Shops like mine are not constrained by processor or disk, as those are easily added on the 6-core models, while on the 4-core models the disk cap is pretty high. When we get down to the brass tacks of adding workloads and truly using IBM i on Power Systems as a consolidation machine, it would be nice to be able to tack on more memory if needed. 64 GB is not enough in my opinion, and I've shared that opinion with Big Blue. They may or may not agree as they have their own metrics on how their systems should be set up and positioned. But it never hurts to give feedback.
Maybe more people should be leveraging their systems like I've been leveraging mine. Maybe they want to. But in the 4-core models, that's tough to do if memory is the biggest constraint. So if you want to have more memory, then currently you need to buy a P10.