| Author |
Message
|
| rbicheno |
 Posted: Tue Jul 05, 2011 6:05 am Post subject: |
 |
|
Apprentice
Joined: 07 Jul 2009
Posts: 43
|
lancelotlinc said:
You have again picked and compared 2 raw numbers, once again those are not useful comparsons without looking at the details i.e. workload and hardware specs. Whilst i was working at IBM I co-authored those reports so i know that testcase is the lightest scenario, its not a realistic workload, probably more nwk/IO bound than CPU, that test is there as an indication as a high watermark for the given platform. If you look in the report appendix you will see the z/OS report was run on an OLD z990 with v slow CPU. I'm sure if you ran on a modern z10 you would see much better numbers.
The performance reports were never produced with the intention of cross platform comparison which you are doing, due to the hardware differences but as an indication of performance for a particular platform on given hardware and to aid sizing/capacity planning. |
|
| Back to top |
|
 |
| lancelotlinc |
| Posted: Tue Jul 05, 2011 6:13 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
I agree with your points about cross-platform comparisons by the reports you mentioned. And excellent work on those reports by the way. Absolutely stellar reporting.
I also like your idea about comparing TCO across platforms being more meaningful. Did you want to post a z/OS TCO calculation ? Then, I'll do the same for Power7 and we can discuss the merits of each.
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| rbicheno |
| Posted: Tue Jul 05, 2011 6:20 am Post subject: |
|
|
Apprentice
Joined: 07 Jul 2009
Posts: 43
|
Alas i have left IBM and no longer have access to that data, and we don't use z/OS where i work now. But note I was disagreeing with your comparison method not your TCO conclusions  |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Jul 05, 2011 6:21 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| lancelotlinc wrote: |
| I agree with your points about cross-platform comparisons by the reports you mentioned. |
Then why use them for cross-platform comparisons?
Define "TCO" before we go any further. What are you including in the cost, and how far through the organisation are you spreading it?
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Tue Jul 05, 2011 6:29 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
Here is what I would consider for TCO:
Computer hardware and programs
Network hardware and software
Server hardware and software
Workstation hardware and software
Installation and integration of hardware and software
Warranties, extended service cost and licenses
Operation expenses
Infrastructure (floor space)
Electricity (for related equipment, cooling, backup power)
Testing costs
Downtime, outage and failure expenses
Information technology personnel
Great point Sir Vitor. Let's define what items should be in our TCO calculation first. Anyone add anything to the list, or nominate any items to remove from the above list?
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Jul 05, 2011 6:31 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
And your answer to my other question, on who's carrying the cost?
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Tue Jul 05, 2011 6:44 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
| Vitor wrote: |
| And your answer to my other question, on who's carrying the cost? |
Ultimately it would boil down to the customer carrying the cost. For example, the retail price of a credit card transaction that credit card networks charge merchants is US$0.55. I would use three years as the length of time to show a return on investment for the new system. Anyone disagree with three years?
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Jul 05, 2011 6:50 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| lancelotlinc wrote: |
| Vitor wrote: |
| And your answer to my other question, on who's carrying the cost? |
Ultimately it would boil down to the customer carrying the cost. For example, the retail price of a credit card transaction that credit card networks charge merchants is US$0.55. I would use three years as the length of time to show a return on investment for the new system. Anyone disagree with three years? |
Apologies for my lack of clarity. I meant system customers. We assume that no-one buys a z/OS sysplex and just runs broker on it. I would consider it unlikely (but stand to be corrected) no-one buys an expensive Power7 box and just runs broker on it. So how much of the costs you mention (aside from the license of WMB itself) accure to WMB?
We'll get at some later point to the question of if WMB is running on a box with other software how that affects performance, given that z/OS and AIX (like most Unix) use different scheduling models for resource allocation.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Tue Jul 05, 2011 6:54 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
I am open to your suggestion. Perhaps half of the total cost would be allocated to WMB. Does this seem reasonable?
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Jul 05, 2011 6:55 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| lancelotlinc wrote: |
| For example, the retail price of a credit card transaction that credit card networks charge merchants is US$0.55. |
Having worked for a number of UK banks, I consider that figure mostly irrelevant. Internally documented cost of a credit card transaction for a major UK credit card provider (which I can't name but has bird roadkill as a logo) was GBP0.06. Cost to the merchant was GBP0.35. One year the cost was recalculated as GBP0.06 but the external price rose to GBP0.40 due to "increased costs". When I enquired what cost, I was told half-seriously "the Chairman's favourite brand of champers has gone up".
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| Vitor |
| Posted: Tue Jul 05, 2011 6:57 am Post subject: |
|
|
Grand High Poobah
Joined: 11 Nov 2005
Posts: 26093
Location: Texas, USA
|
| lancelotlinc wrote: |
| I am open to your suggestion. Perhaps half of the total cost would be allocated to WMB. Does this seem reasonable? |
Hardly. On the average z/OS sysplex WMB is unlikely to be half of the workload. I'd have said nearer 3% - 5% with 10% as an upper band and would welcome shout ins from other z/OS WMB people at this point.
_________________
Honesty is the best policy.
Insanity is the best defence. |
|
| Back to top |
|
|
| sipples |
| Posted: Wed Sep 21, 2011 3:12 am Post subject: |
|
|
Newbie
Joined: 21 Sep 2011
Posts: 7
|
Many people have their favorite baseball (cricket?) teams, beverages, vacation destinations, and platforms. But I think it's important that we stick to current facts even while arguing about our favorites. So here are some facts:
1. Prior to WebSphere Message Broker V7, WebSphere Message Broker for z/OS used 31-bit (2 GB) address spaces, so memory was allocated in 2 GB (or less) portions. That has *never* been a z/OS restriction or limitation: z/OS has always supported 64-bit addressing and huge memory allocations. (As did OS/390 V2R10 which came before z/OS.) If memory was a limitation for you with WMB in the past, that should no longer be the case with WebSphere Message Broker V7 or higher for z/OS. Assuming you've given (that particular) z/OS sufficient physical memory, of course.
2. As noted above, if you compare a smaller system which was state-of-the-art 7+ years ago to a bigger system that is state-of-the-art today, it would be remarkable if the former delivered the same or similar performance and throughput as the latter. Said another way, IBM publishes performance reports, but the expectation is that you'll employ at least basic reading comprehension.
Now, maybe IBM shouldn't be making that assumption. But please do read carefully anyway.
3. IBM is publicly committed to competitive mainframe total costs. In the specific case of WMB for z/OS, here are just a few of IBM's recent price-reducing actions: System z Solution Editions (new WMB workloads are eligible), "Getting Started" and IWP pricing (i.e. mainframe-unique sub-LPAR sub-capacity pricing terms), zAAP enablement for Java compute nodes, ample XML and Java performance work (two notable examples), and z114 pricing. To elaborate on Solution Edition, that's total multi-year *pricing* (hardware, software, and maintenance) -- not only total cost/TCO -- designed to be competitive with the best/realistic non-IBM alternatives.
4. Performance varies, even with patch levels -- and sometimes dramatically. (Example: Java 6.0.0 to 6.0.1 on z/OS. That third digit makes for another big improvement.) From time to time IBM publishes performance reports, and they're often quite perishable. Look for document "swg27007150" to find them. (That document number might change in the future, but I think it's pretty durable.)
5. There are some things WMB for z/OS can do that WMB elsewhere cannot. Said another way, the performance of local in-memory access to z/OS-hosted resources, z/OS Workload Manager exploitation, the VSAM/QSAM nodes, SMF accounting, ARM exploitation, and Parallel Sysplex exploitation, to pick a few examples, is *infinitely* better with WMB z/OS than it is with any other WMB platform.
6. In-memory connection is more efficient, ceteris paribus, for *both* parties to any communication. (More efficient means less CPU consumption.) To pick an example, DB2 has to work a lot harder to service queries and updates from outboard than inboard, other things being equal. That's especially true if you've got personally identifiable information (PII), e.g. credit card data, that you have to encrypt over network hops. Don't forget that important consideration.
7. Your best/perfect WMB implementation could well be some WMB z/OS and some WMB for one or more other platforms. Pause and reflect on that for a moment: WMB actually works very well with WMB.
Please note that I do not speak for IBM. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Wed Sep 21, 2011 4:47 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
Very nice article, thank you for your words of wisdom. I appreciate the vantage point from which you speak.
The proof is in the pudding however. I agree with you that no two IBM performance reports, unless conducted at the same time with the same controls, will be any better than an apple to orange comparison.
That being said, please address the problem of CPU time on z/OS for WMB transactions. Not only do the IBM performance reports indicate that there is a two-to-one difference between Power 7 and z/OS, but in my personal experience I also have actually observed this issue.
Given the same payloads and bar files, in my personal experience at two major household-name insurance companies, a nation-wide bank, and one of the two major credit card networks, I have observed that z/OS processing is by far slower than Power 7 processing.
Why is that?
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
| mqjeff |
| Posted: Wed Sep 21, 2011 4:50 am Post subject: |
|
|
Grand Master
Joined: 25 Jun 2008
Posts: 17447
|
| lancelotlinc wrote: |
I have observed that z/OS processing is by far slower than Power 7 processing.
Why is that? |
My first, random, guess would be that the AIX lpars were extensively tuned for Broker and the power 7 hardware was extensively tuned to support those LPARs, and that the zOS LPARs were tuned for more general purpose work with the corresponding hardware tuning. |
|
| Back to top |
|
|
| lancelotlinc |
| Posted: Wed Sep 21, 2011 4:52 am Post subject: |
|
|
Jedi Knight
Joined: 22 Mar 2010
Posts: 4941
Location: Bloomington, IL USA
|
| mqjeff wrote: |
| lancelotlinc wrote: |
I have observed that z/OS processing is by far slower than Power 7 processing.
Why is that? |
My first, random, guess would be that the AIX lpars were extensively tuned for Broker and the power 7 hardware was extensively tuned to support those LPARs, and that the zOS LPARs were tuned for more general purpose work with the corresponding hardware tuning. |
I have not seen in the WMB Installation Guide any mention of these criteria. What z/OS tuning is needful to achieve par on the performance with other platforms?
_________________
http://leanpub.com/IIB_Tips_and_Tricks
Save $20: Coupon Code: MQSERIES_READER |
|
| Back to top |
|
|
|
|
