This Question is Answered
I believe the goal was to configure the systems as close to the same price as possible. They say PC2100 was the fastest RAM available at the time of testing for the DL585, I don't know what CPUs were available from AMD 6 months ago.
Be skeptical of all benchmarks
Be skeptical of all benchmarks
This is an interesting thread.
I'll start with : "Once and for all: HP DL585 vs. IBM x460 (or x366) performance". Paulo I assume you are joking ...... once for all ? Come on..... this will go on forever .......
Don, I wouldn't be so negative. Assuming you are using the same CPU's the performance would be pretty much the same (i.e. the same) with the only difference that the 366 is slightly cheaper than the 460 being it limited to 4-way.
Lee, I think it depends. If you ask me "would you put a 4-way exchange server on ESX" I would probably say no and I would buy a dedicated server. If on the other hand you have a batch workload that runs during the night and would like it to run in the shortest time possible wouldn't you creat a 4-way vm that runs at 100% of full utilization say from 2 AM to 3:30 AM ? I would. Well if we end up with a 4-way vm that performs like a single or dual cpu's physical server than it's another story ....... but if we assume it will run reasonably welli.e. some 10-20% of performance penalty compared to a physical 4-way........... that could be a good trade off.
For the records Whitefield does not exist anymore. This has been replaced with Tigerton which implements what Intel refe to as Direct Attach Interface (or something like that). That is a 4-core CPU's with each socket directly connected to the chipset. This will happen in 2007. Intel claims it will be much better than the Opteron but ..... what do you expect them to say ? Please don't start a battle on this ....... it's worthless.
Simon, you hit a good point here. Unfortunately that is the boundary between the zSeries mainframes or the AS/400 and the Intel "nightmare". Yes we could do what we do on the mainframe in terms of testing but that would limit the list of the certified servers to 2 or 3, they would be listed 6 months after the release of the HW/SW and those systems would cost 4x what they cost today. You guys are going to pay for that and this wouldn't work in this market because everyone is still concerned about the $$$$$ and who costs (reaonably) less would win. I am not talking about the + or - 5/10% everyone is ready to pay for wuality .... I am talking about 4x the current price.
I agree.... see above.
Sorry about that but VMware does not allow us to publish it before the beta2 release is out.
what am I supposed to say here ? I will go for a standard "no comment". Joking aside ...... bringing a product to market costs money and you have to have a compelling reason to do that (i.e. you have a product hole). Given the fact that the x366 and the 4-way Opteron overlaps for some 99% of the aspects (or something around that.......). Well of course if, for some reasons, you ALL start buying Opteron only kits then there is a compelling reasons....... Notice that although AMD has made HUGE in-raods in the x86 market space Intel still have the crown. If you also consider that the in-roads are made by HP that is no longer "promoting" Intel based kits in the 4-way space ..........
But this is a complex (marketing) matter........ I wouldn't want to be in the IBM room where they take these decisions............
Sun has stated they would have one. I haven't heard anything from HP.
The test was run on two similarly priced systems and PC2100 was the memory technology that that server supported.
Probably never.... but since the bid price might change based on a customer/situation basis .... we opted for the web price. You have to start somewhere. Also we have taken into account web price of the x366 as well of course.
No. That was not the case. The 366 was fully up to date and the DL585 not leading edge only because this was meant to be a price/perf play.
Are you implying that we hacked the HP system for it to perform badly ? be assured ..... we still have a soul..... I think ... 
Excellent thread .....
Massimo.
I'll start with : "Once and for all: HP DL585 vs. IBM x460 (or x366) performance". Paulo I assume you are joking ...... once for all ? Come on..... this will go on forever .......
a comparison between a 4 way x460 and a 4 way DL585 is not very >useful, I image the DL585 would win the majority of benchmarks
Don, I wouldn't be so negative. Assuming you are using the same CPU's the performance would be pretty much the same (i.e. the same) with the only difference that the 366 is slightly cheaper than the 460 being it limited to 4-way.
OH MY GOSH!
Brian, I agree LOL :-D why the hell would you run 4way VM's?
Lee, I think it depends. If you ask me "would you put a 4-way exchange server on ESX" I would probably say no and I would buy a dedicated server. If on the other hand you have a batch workload that runs during the night and would like it to run in the shortest time possible wouldn't you creat a 4-way vm that runs at 100% of full utilization say from 2 AM to 3:30 AM ? I would. Well if we end up with a 4-way vm that performs like a single or dual cpu's physical server than it's another story ....... but if we assume it will run reasonably welli.e. some 10-20% of performance penalty compared to a physical 4-way........... that could be a good trade off.
Yes the global scheduler will help
You need to be careful here. Many see DRS as a distributed vmkernel scheduler (VMware marketing charts lead towards this concept) but ....... guys ....... this technology, AS OF TODAY, has nothing to do with a distributed scheduler. It's like having a dedicated operator doing vmotion based on some colours on the screen....... Also the vmkernel scheduler has a timescale in the range of milliseconds, DRS has a timescale that is minutes/hours. It will improve of course...... but DRS is not the tool that will allow you to use low-end hardware because it will treat those all ogether as if they were a single high-end server.... Back to the main subject as mentioned its swings and roundabout, the >460 doesnt even work in full mode yet (full RAM etc), The 366 I dont think >is supported as yet (massimo correct me if I am wrong),
The 366 and 460 have just been listed as supported with the dual-cores. There is a bug in the 2.5.2 that prevents the usage of 64GB of memory. VMware is supposed to fix it in 2.5.3. I hope with Dempsey and Whitfield to see more improvements
For the records Whitefield does not exist anymore. This has been replaced with Tigerton which implements what Intel refe to as Direct Attach Interface (or something like that). That is a 4-core CPU's with each socket directly connected to the chipset. This will happen in 2007. Intel claims it will be much better than the Opteron but ..... what do you expect them to say ? Please don't start a battle on this ....... it's worthless.
If we look back inwards and something I have been pursuing with a >passion is viewing of the certification process that VMware asks the OEM >to go through, the quantity undertaken and results of this have to be >made public.
I would feel alot more comfortable with knowing that to be certified on >2.5.x IBM/HP/DELL have tested say 50-100 of these boxes and in >different load scenarios.
To me at the moment this is not happening (Massimo, again here to be >corrected) if it is, would we be seeing the issues that have been faced >and continue to be faced and having to wait for another patch for fixes >etc.
I would feel alot more comfortable with knowing that to be certified on >2.5.x IBM/HP/DELL have tested say 50-100 of these boxes and in >different load scenarios.
To me at the moment this is not happening (Massimo, again here to be >corrected) if it is, would we be seeing the issues that have been faced >and continue to be faced and having to wait for another patch for fixes >etc.
Simon, you hit a good point here. Unfortunately that is the boundary between the zSeries mainframes or the AS/400 and the Intel "nightmare". Yes we could do what we do on the mainframe in terms of testing but that would limit the list of the certified servers to 2 or 3, they would be listed 6 months after the release of the HW/SW and those systems would cost 4x what they cost today. You guys are going to pay for that and this wouldn't work in this market because everyone is still concerned about the $$$$$ and who costs (reaonably) less would win. I am not talking about the + or - 5/10% everyone is ready to pay for wuality .... I am talking about 4x the current price.
I agree. It concerns me that server hardware is being introduced to >market before robust testing is carried out. In defence of IBM/HP/Dell its >going to be a very difficult task to cover every avenue and combination. >When would be the right time to release something?
I agree.... see above.
I'm dissappointed that IBM has not released the Redbook on the x460 by >now. I'd thought it would have been something that came out with the >server, not six to eight months afterwards. I'm going to be implementing >four of these beasts without one, not something I feel comfortable about
Sorry about that but VMware does not allow us to publish it before the beta2 release is out.
If only IBM had an AMD server comparable to the DL585, I'm sure many >IBM shops would have a definite answer...
what am I supposed to say here ? I will go for a standard "no comment". Joking aside ...... bringing a product to market costs money and you have to have a compelling reason to do that (i.e. you have a product hole). Given the fact that the x366 and the 4-way Opteron overlaps for some 99% of the aspects (or something around that.......). Well of course if, for some reasons, you ALL start buying Opteron only kits then there is a compelling reasons....... Notice that although AMD has made HUGE in-raods in the x86 market space Intel still have the crown. If you also consider that the in-roads are made by HP that is no longer "promoting" Intel based kits in the 4-way space .......... But this is a complex (marketing) matter........ I wouldn't want to be in the IBM room where they take these decisions............
does anyone known, if HP or SUN will bring an 8 CPU Dual Core Operton
System in 2006?
System in 2006?
Sun has stated they would have one. I haven't heard anything from HP.
Also while the IBM got 3.66Ghz CPU's (Which IIRC were the FASTEST >Intel's at the time) the HP box only got 848's and not 852's which are 15->20% faster on core clock.
The test was run on two similarly priced systems and PC2100 was the memory technology that that server supported.
When was the last time an enterprise customer paid web price for a >DL585?
Probably never.... but since the bid price might change based on a customer/situation basis .... we opted for the web price. You have to start somewhere. Also we have taken into account web price of the x366 as well of course.
Also IBM supplied the DL585 so it may have been sitting on a shelf for a >few months before being shipped to the testing center, in which case the >PC-2100 was probably the best that they could do, but that was not the >case on the test date, I bet the x366 was fully up to date.
No. That was not the case. The 366 was fully up to date and the DL585 not leading edge only because this was meant to be a price/perf play.
At the end of the day a vendor paid for test with gear supplied from one >vendor is not a reliable test IMHO.
Now if someone like ZD Labs procured two systems to a price point and >ran a similar test I would have a LOT more faith in it.
Now if someone like ZD Labs procured two systems to a price point and >ran a similar test I would have a LOT more faith in it.
Are you implying that we hacked the HP system for it to perform badly ?
be assured ..... we still have a soul..... I think ... Excellent thread .....
Massimo.
This is an interesting thread.
I'll start with : "Once and for all: HP DL585 vs. IBM x460 (or x366) performance". Paulo I assume you are joking ...... once for all ? Come on..... this will go on forever .......
I'll start with : "Once and for all: HP DL585 vs. IBM x460 (or x366) performance". Paulo I assume you are joking ...... once for all ? Come on..... this will go on forever .......
Thanks, Massimo, for the reality check.
Despite nobody having come forward with a comparison of the two systems with dual-core CPUs, a lot of useful information has been gathered. Your post brings very interesting points, and I'd pick the one I find the most important: in the end of the day, it's all about money and marketshare. Or, as you put it: "bringing a product to market costs money and you have to have a compelling reason to do that (...) the x366 and the 4-way Opteron overlaps for some 99% of the aspects (...)"So, in order for this not to just keep going forever, I will be awarding points by the end of the week. Let's then go on with the rest of our lives - at least until someone actually has the opportunity to do an apples-to-apples comparison of the two little beasts...
Excellent thread .....
Much better than I expected!
Congratulations to all, and thank you!Paulo
At the end of the day a vendor paid for test with
gear supplied from one >vendor is not a reliable testIMHO.
Now if someone like ZD Labs procured two systems to
a price point and >ran a similar test I would have aLOT more faith in it.
Are you implying that we hacked the HP system for it
to perform badly ?
be assured ..... we still havea soul..... I think ...
No I am not implying a delibrate "Hack" of the DL585 to slow it down.
However vendor paid for benchmarks have a history of being favourable to whoever paid.
