* Posts by Mr Nelson

6 posts • joined 27 Mar 2013

Ellison aims his first Oracle 'mainframe' at Big Blue

Mr Nelson

Re: Ibm marketing...

Kebabbert, Nice try but your whole premise is wrong. When companies evaluate servers they don't ask: "We need to get the highest SAP performance for 8-socket servers." Nobody buys Unix servers based on processor=sockets because that doesn't mean or do anything for them. It's a useless metric. If Oracle believes customers will do that they will be sorely disappointed and will continue to lose market share at a rapid pace as they have been ever since they bought Sun. Only in some cases in the x86 world people look at the socket level to make server decisions. For example, when VMware or Microsoft license counts are at stake. But even in those cases they must be very careful about how many cores the servers have, like in the case of also needing Oracle database and Oracle RAC licenses which could cost them a fortune if they ignore the core count.

When looking for SAP servers what customers ask is "What server can meet my performance requirements?" If for example they need 200,000 SAPS, they couldn't care less if it has 8 or 16 or 40 processors, what they care about is having a server that can handle 200K SAPS. Assuming the customer wants all those SAPS in a single server (rarely the case) then the Power 760 wouldn't be the server because it maxes out at around 140k SAPS. They probably use the Power 770 or if the customer wants to buy based on published SAP benchmarks then the Power 780. The Power 780 supports 311,720 SAPS with just 96 cores installed, but unlike the T5-8 the customer doesn't have to purchase the entire server with 12 processors and 96 cores but just what it needs to support those 200k SAPS; that is, 62 cores (200,000 / 3,247 SAPS per core). That's how SAP servers are sized, not for SAP application costs but to know how much server capacity they need to run the workloads. Do you know how many Power 780 processors the customer would need to purchase to support 200k SAPS? Eight. Yet another blow to your performance premise. And like I said in my previous comment, unlike the T5-8 they don't even have to buy all the CPU and memory capacity at once because of comprehensive Capacity on Demand features for cores and memory. So they'll save money on hardware and software by paying for capacity as they grow. And they can grow up to 311,720 SAPS without changing hardware if they choose or need to. No customer is going to come back to IBM and say: "I'm so disappointed you didn't propose a server that only supports 8 Processors and no more because that was the most critical requirement we had. We are disqualifying you for not meeting that most important, non-negotiable requirement." You would love that reaction, but no, it's not real life.

Mr Nelson

Phil 4,

It seems you greatly exaggerated by quite a bit the cost of the Power 760 with the CPU and memory capacity used in the SAP benchmark (48 CPU cores and 1 TB of memory). The cost is in the neighborhood of $470k and that includes 3-Year/24x7 support for hardware and software. That cost figure also includes PowerVM virtualization software. I suspect that the support costs you listed for the T5-8 is not even 3-Year 24x7 support. And I have already proven that Oracle DB software would cost a lot more for the T5-8 due to the need for 16 additional DB licenses at $47,500 for each license plus $10,450 software maintenance Per Year. For a 5-Year period that would be $99,750 per license, or $1,596,000 total just for the 16 additional licenses. Ouch! It seems your 3x cost advantage claim has evaporated and turned into a big disadvantage, more like in the neighborhood of 2x in favor of the Power 760 when database licenses are included; as it should because databases were used in the benchmark.

Furthermore, the Power 760 has enterprise features the T5-8 could only dream off, such as CPU Capacity on Demand, that is CPU as in cores, so that IT doesn't have to pay for all CPU cores and software licenses from the start (including for Oracle DB), if they don't need to for their workload's size. Then they can activate and pay for only the additional CPU capacity they need as the workload grows, and do so dynamically and without downtime. What a concept. That's unlike the T5-8 where you are stuck with the entire box and required licenses all at once. The Power 760 also has Active Memory Expansion (yet another feature the SPARC line can only wish to have), for real-time compression/decompression of memory so that less physical memory is needed for SAP workloads. Because it actually works, Active Memory Expansion is officially supported by SAP (see SAP Note 1464605 - Active Memory Expansion (AME)).

For those who don't have access to this SAP Note, here's an extract of what the note says:

"The AIX feature Active Memory Expansion for IBM POWER systems is a new technology for expanding a system's effective memory capacity. It is available starting with POWER7. Active Memory Expansion employs memory compression technology to transparently compress in-memory data, allowing more data to be placed into memory and thus expanding the memory capacity of POWER systems. Utilizing Active Memory Expansion can improve system utilization and increase a system's throughput."

Mr Nelson
FAIL

Re: Ibm marketing...

No, Kebbabert, you are the one distorting the real facts. You want to hang on to a claim with no real value to anyone. Customers do buy per core and that's what matters when choosing servers. It matters for licensing and it matters for server costs. The only reason you hide behind useless processor claims is because those SPARC boxes have 16 cores per processor, and so the number of cores is what makes the difference in your performance numbers not the number of CPUs. So to you a 1-processor box with 8 cores should be equal to a 1-processor box with 16 cores so that poor Oracle can have its useless performance claims validated.

Mr Nelson

Funny that you mention it. Let's take a look at your "real" cost examples. For argument's sake, let's assume for a second that your Power 760 cost is accurate, then we have a total cost advantage for someone who wants to run Oracle databases on the IBM box. I guess your forgot that people need to pay for database licenses,? But you didn't include that in your cost estimate. Since it only has 48 Oracle licenses based on Oracle's Per Processor policy and the T5-8 has 64 (128 * 0.5), that's 16 additional Oracle licenses for the T5-8 at a whopping $760,000 more just for the extra licenses. The you add 5-Year year maintenance to that and you end up paying well over a million dollars more for the T5-8.

Now let's look at TPC-C costs. If you go to www.tpc.org and look at at both the IBM and Oracle documentation for the benchmark, you'll see some issues pop up. When including database costs in its TPC-C benchmarks, IBM uses enterprise type of database license and maintenance costs. Oracle is pricing its Oracle DB EE and the Partitioning support as a 3 year TERM license so 50% of 'full' price and then then applying the T5's 0.5 core factor. SW support is "server incident" for $2300/year/server (which allows only web-based submission of up to 10 incidents per year). This all saves Oracle in the neighborhood of $4 Million on the list price as compared to using Perpetual SW licenses and Premier Support for the DB. In addition, where's the Maintenance/Support in the pricing for the server or storage? Who in their right mind would run an enterprise system with web-based server incident support, and for just 10 incidents per year and no hardware maintenance? Oracle has the right to use whatever pricing method they choose to use for a benchmark, but if you want to claim apples-to-apples pricing then use comparable configurations.

Mr Nelson
Thumb Down

Re: Ibm marketing...

You are simply wrong. The number of cores affects the cost of the hardware, not just the number of CPUs. Do you think you'll pay the same for processors regardless of the core count? A 2-processor / 8 core box doesn't cost the same as a 2-processor / 20-core box. That applies to hardware as well as software, especially for Oracle licenses. When Oracle is benchmarketing, processors equal sockets but when licensing Per Processor then processors are cores. Nicely done. Try buying Oracle licenses based on processors = sockets. Sizings for applications like SAP are done taking the SAPS per Core performance as the critical factor, not processors/sockets. If Oracle wants people to think of performance based just on processors they should license their products the same way and then they can argue about the merits of whole CPUs.

Mr Nelson

Larry Ellison is it again claiming he surpassed IBM Power's performance. Don't believe that for a second. Here are the facts. Very conveniently Oracle compares server performance based on number of processors, a.k.a, chips or sockets. Processors/Chips/sockets are not a real measurement of server performance for the simple reason that different processors have different core counts. For example, a 2-processor server from one vendor may have a total of 16 cores while another may have 32 cores. That's precisely the case with Oracle's T5 family. Most of Oracle's so called World Records are based on thsi premise. For example, for SAP performance Oracle compares the 8-processor T5-8 against the benchmark result of the 8-processor Power 760. Apples to apples, Oracle says, but the T5-8 had 128 cores and the Power 760 had 48 cores. Clearly not a fair comparison. In reality the Power 760 beats the T5-8 in Per-Core performance: 2,900 SAPS/core vs 1,726 SAPS/core for the T5-8. It seems like Power is still the performance king and by a good margin. Oracle does the same thing in almost of benchmarks it announced during the announcement.

Biting the hand that feeds IT © 1998–2019