Entry Server Bang For The Buck, IBM i Versus Red Hat Linux
November 11, 2019 Timothy Prickett Morgan
In last week’s issue, we did a competitive analysis of the entry, single-socket Power S914 machines running IBM i against Dell PowerEdge servers using various Intel Xeon processors as well as an AMD Epyc chip running a Windows Server and SQL Server stack from Microsoft. This week, and particularly in the wake of IBM’s recent acquisition of Red Hat, we are looking at how entry IBM i platforms rate in terms of cost and performance against X86 machines running a Linux stack and an appropriate open source relational database that has enterprise support.
Just as a recap from last week’s story, the IBM i matchup against Windows Server systems were encouraging in that very small configurations of the Power Systems machine running IBM i were less expensive per unit of online transaction processing performance as well as per user. However, on slightly larger configurations of single socket machines, thanks mostly to the very high cost per core of the IBM i operating system and its integrated middleware and database as you move from the P05 to P10 software tiers on the Power S914, the capital outlay can get very large at list price for the Power iron, and the software gets very pricey, too. The only thing that keeps the IBM i platform in the running is the substantially higher performance per core that the Power9 chip offers on machines with four, six, or eight cores.
Such processors are fairly modest by 2019 standards, by the way, when a high-end chip has 24, 28, 32, or now 64 cores, and even mainstream ones have 12, 16, or 18 cores. If you want to see the rationale of the hardware configurations that we ginned up for the comparisons, we suggest that you review the story from last week. Suffice it to say, we tried to get machines with roughly the same core counts and configuration across the Power and X86 machines, and generally, the X86 cores for these classes of single socket servers do a lot less work.
With the Linux to IBM i comparisons, the big difference is that software is priced per core perpetually on the Power iron but it is priced per socket per year on the Linux machines. As it turns out, the Linux software stack at standard feature and support levels that we set up on the four-core X86 machines are more expensive than the IBM i stack, but even if you move to the premium features and support on the six-core and eight-core machines, the Linux stack is still significantly less expensive than IBM i.
Let’s back up for a second and talk about the comparisons. On the Power S914 system, the machines are configured with IBM i 7.4, which includes the PowerVM hypervisor, Java middleware, and the Db2 for i relational database.
On the Linux setups, we started with the Red Hat Virtualization hypervisor, which is a commercial implementation of the KVM hypervisor. The four-core machine has the Standard version of RHV, which costs $999 per year, and the six-core and eight-core versions have the Premium edition, which costs $1,499. We priced it out over three years, which generally where perpetual and subscription license costs cross over. On top of that, we added Red Hat Enterprise Linux for Virtual Datacenters, which is the version of RHEL that can be highly virtualized with lots of logical partitions on a machine. On the four-core machine, the RHEL we chose (which is a lot more expensive than the $349 for a base, bare metal RHEL license) has a standard support, which costs $2,499 per year per socket and on the heftier X86 boxes with six or eight cores, it has premium support, which costs $3,999 per socket per year. For middleware, we added JBoss Enterprise Application Platform Standard to the baby box and Premium on he two bigger ones; that is $8,000 per socket per year on the former and $12,000 per socket per years on the latter. That Java middleware is very, very pricey. Like a lot more than the relational database.
For the database, we went with the go-to open source database these days, which is the EnterpriseDB Advanced Server implementation of the PostgreSQL database. The EnterpriseDB Basic license costs $1,500 per socket year plus $1,000 per year for Standard support; the Premium variant of the database costs $5,000 per socket per year plus $3,000 per socket per year for Premium support.
Here are the feeds and speeds of the IBM i and Linux setups:
This is, admittedly, the Cadillac version of a Linux stack. You could use the Tomcat application server, cheaper versions of RHEL and even EnterpriseDB. You could spend the same money and run it on a much more powerful single-socket server and spread that software cost across more cores in a socket, thus delivering better bang for the buck. But for a reasonably comparable stack of software with reasonably similar software, the open source software stack over three years running on X86 iron has a lower capital outlay on six-core and eight-core machines, and costs considerably more on a four-core machine. Yup. You read that right. And when you do the math on a cost per user – allocating 2,100 CPWs per user as we did with the Windows Server stacks – the Linux stacks as I have configured them are more expensive per user than the IBM stack across the board. As you can see from the monster table above, the hypervisor is a little more or a little less expensive that the VMware ESXi hypervisor we put on the Windows Server stack. The RHEL operating system is not that pricey, but JBoss is, and EnterpriseDB is in the same ballpark as SQL Server, give or take.
Take a look:
Peculiar, isn’t it? Linux has this perception of being inexpensive. I realize this is at list price, and there is a much better chance of getting a discount from Red Hat on a Linux stack than from Big Blue for the IBM i stack. Or, maybe that was never true, and maybe it is certainly not true anymore. If Windows Server and Linux are now legacy platforms, which at this point they are, they have to charge a premium for functionality as IBM has had to do for decades with OS/400 and IBM i to keep the whole operation running.
As I said last week, I am still working to get comparisons for the Power S924 and Power E980 together that do the same thing. It is very hard to get server pricing information, so if you know something about pricing on these machines, don’t be shy and share.
One last thought before I go. The most likely scenario for IBM i shops now that Big Blue has acquired Red Hat is not an either/or situation, but an and scenario. It seems far more likely that IBM i shops wishing to simplify their infrastructure will be able to load up the OpenShift Kubernetes container management system on Linux partitions on their power machines and run it side-by-side with their IBM i database environments running in IBM i partitions. We will take a look at what those IBM i and OpenShift situations might look like, because this is one of the reasons why Big Blue bought Red Hat in the first place. There is no reason OpenShift has to run on X86 servers when it runs perfectly well on Power.
RELATED STORIES
Entry Server Bang For The Buck, IBM i Versus Windows Server
While I find this analysis sound and it is good information, I find that it still doesn’t address the core cost problem with IBM i. The problem isn’t the cost to buy your entry level server. The problem is that you have to buy an entry level server just to get started. You’ve already lost the mindshare before this cost equation even becomes a factor. I can work out a proof of concept for an idea and get it working reliably on a small scale using spare parts I dug out of my garage in most cases. Even for a company, you might throw up a virtual CentOS (basically free Red Hat) linux machine and tinker before you decide to start some new project with full support. At that point, you need to decide if its worth your time and effort to figure out how to make that work on a new platform. You also need to do that on a platform that is a great unknown to almost everyone else that uses computing tools on a regular basis. It’s just not going to happen.
Wow, this is so off the mark I don’t even know where to start. You have to look at what you are trying to deliver to define the cost. Yes IBMI comes with virtualization, but you won’t use it in this space, so it’s wrong to add it to the price of a comeptative solution. Likewise with the other application adds. You added a bunch of licensing on the Linux side and left out a bunch on the IBMi side. It’s easy to inflate costs when you are just trying to prove the point you want without actually defining the solution. This comparison is disingenuous at best and in reality, just plain wrong.
TPM, you have a difficult task trying to compare IBM i with an x86 solution. It is difficult even if it were IBM i to AIX for example. With IBM i being so integrated and optimized, you can’t compare 1 x86 system with 8 cores to 1 x POWER9 server with 8 cores. A shop simply won’t deploy a single x86 system to run an enterprise application from Infor for example. Whereas, we all know and accept it is standard to deploy a single IBM i system to run an entire business.
Further, if a 4-core POWER9 S914 were available with more than 64 GB RAM and more storage capacity, as IBM intentionally limits this 4-core model, the smallest available system would probably be the best seller in the lineup …. of course, I just gave everyone the reason as to why IBM “constrains” their smallest server.
Though I found the article interesting, it simply isn’t useful except to perpetuate a myth that clients should compare equal numbers of sockets and cores between Intel and IBM Power. Something with AIX is anywhere from 4 – 10X fewer cores and with IBM i, even fewer resources will be required.
Keep ’em coming though as not every article can be a home run.