Quicksearch
Disclaimer
The individual owning this blog works for Oracle in Germany. The opinions expressed here are his own, are not necessarily reviewed in advance by anyone but the individual author, and neither Oracle nor any other party necessarily agrees with them.
Case design
Thursday, December 10. 2009
While preparing a presentation, i just found a good example, why i prefer the Sun case design over the one of HP.
This is an HP DL785G6 from the top (this one is from an HP video):
This is the Sun X4640 from the top:
This is an HP DL785G6 from the top (this one is from an HP video):
This is the Sun X4640 from the top:
Comments
Display comments as
(Linear | Threaded)
Nice bullshit.
The video: http://www.youtube.com/watch?v=UpJNj4xJmxA
Question: Will your 4660 work online while a processor is defectice? I guess not. SUN have to shutdown the hole box. You can laugh about the case design a long time, i think as long as the 4640 boot up. Laugh about the case, we laugh about your design failures.
The video: http://www.youtube.com/watch?v=UpJNj4xJmxA
Question: Will your 4660 work online while a processor is defectice? I guess not. SUN have to shutdown the hole box. You can laugh about the case design a long time, i think as long as the 4640 boot up. Laugh about the case, we laugh about your design failures.
I agree, I think the DL785G6 is a better design than X4640, because of more PCI-e slots. The extra 3U of space difference is not that much when your racks can't do high power density anyways. The datacenter these days are power density limited, not space limited.
DL785G6 also is cheaper, starting at $10K for the CTO chassis.
Unfortunately X4640 arrived too late, and will be obsolete by Opteron 6100/Nehalem-EX shortly. So you have to expect a 50% depreciation by Q2 of 2010 when AMD and Intel brings 8-12 core CPUs onto the market.
So, neither the DL785G6 or the X4640 is a great buy.
DL785G6 also is cheaper, starting at $10K for the CTO chassis.
Unfortunately X4640 arrived too late, and will be obsolete by Opteron 6100/Nehalem-EX shortly. So you have to expect a 50% depreciation by Q2 of 2010 when AMD and Intel brings 8-12 core CPUs onto the market.
So, neither the DL785G6 or the X4640 is a great buy.
Most x86 boxes drop when a processor fails anyway, even if the OS supports it 
Rather have the same power in 4RU than 7RU in the HP one.
Rather have less power usage too
Rather have the same power in 4RU than 7RU in the HP one.
Rather have less power usage too
> Question: Will your 4660 work online while a processor is defectice? I guess not. SUN have to shutdown the hole box.
Answer: In this case your DL785 will not work either. And in addition, HP box is bigger, heavier, slower, wastes more energy... and it's ugly.
Answer: In this case your DL785 will not work either. And in addition, HP box is bigger, heavier, slower, wastes more energy... and it's ugly.
>> HP box is bigger, heavier, slower, wastes more >> energy... and it's ugly.
And your old, totally out-dated data center with crappy, out-of-date, unregulated CRACs waste multiples of power than the difference between the x4640 and DL785 servers. Clean up and optimize first your data center infrastructure, before you speak about energy efficiency.
Kind regards
Tschokko
And your old, totally out-dated data center with crappy, out-of-date, unregulated CRACs waste multiples of power than the difference between the x4640 and DL785 servers. Clean up and optimize first your data center infrastructure, before you speak about energy efficiency.
Kind regards
Tschokko
Oh ... when you want to talk about design failures:
How did you design this cool feature that allows you to deliver 8280 SD-User, 45.350 SAPS with Windows and 8.022 SD-User, 43.880 SAPS,Certificate 2009052 with Linux at 2.8 GHz? Given the fact, that the Sun server delivers 10.000 SD-User, 55.070 SAPS at 2.6 GHz this was surely a challenge ! Is Solaris that superior to Linux or Windows? Then you are selling the wrong operating system. Or is there some epic bottleneck in your cooooool system I'm fine with both outcomes. And don't start with "different DB" or "different memory size". Any person knowledgeable about the SAP-SD benchmark should know that the impact is quite small.
7 Rack units instead of 4 ... it was surely a challenge to design such a big case And all this wiring there ....
I would like to talk about power consumption here ... but sorry ... HP power advisor is just Windows only
Before you brag about any feature, you should just get your basics correct, like delivering a server with decent performance
How did you design this cool feature that allows you to deliver 8280 SD-User, 45.350 SAPS with Windows and 8.022 SD-User, 43.880 SAPS,Certificate 2009052 with Linux at 2.8 GHz? Given the fact, that the Sun server delivers 10.000 SD-User, 55.070 SAPS at 2.6 GHz this was surely a challenge ! Is Solaris that superior to Linux or Windows? Then you are selling the wrong operating system. Or is there some epic bottleneck in your cooooool system
I'm fine with both outcomes. And don't start with "different DB" or "different memory size". Any person knowledgeable about the SAP-SD benchmark should know that the impact is quite small.7 Rack units instead of 4 ... it was surely a challenge to design such a big case
And all this wiring there .... I would like to talk about power consumption here ... but sorry ... HP power advisor is just Windows only
Before you brag about any feature, you should just get your basics correct, like delivering a server with decent performance
Joerg:
It is amazing what a Sun employee would do to pump a Sun server.
Last time, we already discussed that the SAP benchmark is rigged because the X4640 had double the ram and ran Solaris. Not because the X4640 is faster than DL785. Those two servers are interchangeable. DL785 gives you more pci-e slots and front loading Processor blades. X4640 gives you 3U of space savings. That's all there is to it. If your space costs more than power(like in California) and you can drive 10kW+ per rack then you would probably consider the X4640. If you are power density restrained, the DL785 would be a better choice.
Both will be obsolete in 3 months. Oh well.
It is amazing what a Sun employee would do to pump a Sun server.
Last time, we already discussed that the SAP benchmark is rigged because the X4640 had double the ram and ran Solaris. Not because the X4640 is faster than DL785. Those two servers are interchangeable. DL785 gives you more pci-e slots and front loading Processor blades. X4640 gives you 3U of space savings. That's all there is to it. If your space costs more than power(like in California) and you can drive 10kW+ per rack then you would probably consider the X4640. If you are power density restrained, the DL785 would be a better choice.
Both will be obsolete in 3 months. Oh well.
It's interesting how awkward you try to make a point, but you just show everytime that you look from your singular view. But: This is getting boring. I'm asking myself everytime a mail indicates a comment from you what's may be your incentive to answer here. Do you just wait that i write one time, that you are right? Well ... then try to make a point that is correct from a multitude of views.
Please stop to see your perspective as the only valid one. Other people have different mindset, different requirements and so on. And they are as correct as yours. And please stop to consider everybode else as morons.
At first: Some expert of the SAP-SD benchmark will gladly explain you, that having more memory doesn't help you in SAP-SD. In contrary: Having more memory can hurt you, if your memory subsystem has to downclock with large DIMMS (that's a classic with Power system clocks down to 400 MHz in a large memory system). This leads to the situation that some people don't get the performance indicated by benchmarks, because they need more memory for their workload.
Just in case you didn't understood this: My comment was meant to show that HP has a problem anyway ... either they have to admit, that Linux or Windows aren't feasible choices for the DL785 (which would be that nice for them) or they have to admit that the HP system is somehwhere bottlenecked.
And by the way: The servers aren't interchangeable ... at 8 sockets you have plenty ways to differentiate. Starting at points where you connect the i/o subsystems to the exact Hypertransport topology to the frequency of the Hypertransport.
I don't think it's problematic to cool 14 kilowatts ... you can't to it with a standard cooling but there are nice an relatively cheap cooling doors available for racks with up to 35 kw. At 14 kw the air should get out of the rack as almost as cold as it got in ...
And maybe rack space is cheap in central somewhere... a rack in California, Tokyo, Munich isn't ... we talked about that a while ago, that you can't compare your hoster RU with the internal costs of a RU for customer in a larger city. But you don't seem to get this as well as other things.
Furthermore i doubt, if the amount of slots is really relevant outside of niches. At the moment many people try to consolidate fabrics, try to reduce cabling. Thus they opt for 10 GBe, opt for 4/8 GB FC or Infiniband. 4 Slots would be sufficient in this case giving you 4 10 GBe interfaces and 4 4/8 GB FC.
Maybe the X4640 update is late, but at first the upcoming systems have to proof that they are delivering the same or better performance with 4 cores than at other systems with 8 cores. I'm sure they will be pretty competitve as the new it the enemy of the old, but i will just wait and see ...
Please stop to see your perspective as the only valid one. Other people have different mindset, different requirements and so on. And they are as correct as yours. And please stop to consider everybode else as morons.
At first: Some expert of the SAP-SD benchmark will gladly explain you, that having more memory doesn't help you in SAP-SD. In contrary: Having more memory can hurt you, if your memory subsystem has to downclock with large DIMMS (that's a classic with Power system clocks down to 400 MHz in a large memory system). This leads to the situation that some people don't get the performance indicated by benchmarks, because they need more memory for their workload.
Just in case you didn't understood this: My comment was meant to show that HP has a problem anyway ... either they have to admit, that Linux or Windows aren't feasible choices for the DL785 (which would be that nice for them) or they have to admit that the HP system is somehwhere bottlenecked.
And by the way: The servers aren't interchangeable ... at 8 sockets you have plenty ways to differentiate. Starting at points where you connect the i/o subsystems to the exact Hypertransport topology to the frequency of the Hypertransport.
I don't think it's problematic to cool 14 kilowatts ... you can't to it with a standard cooling but there are nice an relatively cheap cooling doors available for racks with up to 35 kw. At 14 kw the air should get out of the rack as almost as cold as it got in ...
And maybe rack space is cheap in central somewhere... a rack in California, Tokyo, Munich isn't ... we talked about that a while ago, that you can't compare your hoster RU with the internal costs of a RU for customer in a larger city. But you don't seem to get this as well as other things.
Furthermore i doubt, if the amount of slots is really relevant outside of niches. At the moment many people try to consolidate fabrics, try to reduce cabling. Thus they opt for 10 GBe, opt for 4/8 GB FC or Infiniband. 4 Slots would be sufficient in this case giving you 4 10 GBe interfaces and 4 4/8 GB FC.
Maybe the X4640 update is late, but at first the upcoming systems have to proof that they are delivering the same or better performance with 4 cores than at other systems with 8 cores. I'm sure they will be pretty competitve as the new it the enemy of the old, but i will just wait and see ...
Personally, I think that you resent my presence here because I have never said anything nice about Sun's stuff other than ZFS and Solaris. However, your resentment toward me is blinding your own reasoning.
I wouldn't have said anything more correct from "multiple of views". I said, if you have space issues but can push high power density at 10kW or more, you should consider X4640. If you don't have space issues or only can push 6kW a rack, DL785 is a better choice.
As to the rigged SAP benchmark, Sun still refused to run Linux or reduce the memory down to 128GB to take away all the advantages it wants to give itself to showcase the X4640. As long as the two software configuration and memory configurations are not the same, you cannot say X4640 is faster than DL785G6. It is simply not true.
If you want to prove that X4640 is faster, take half the memory out, and run Linux or Windows, then compare that to the score posted by DL785. Until then, you don't have any data to support your thesis that X4640 is faster than DL785.
You know what I find annoying about you? You don't seem to be the engineer who look at hard cold numbers. You let your employment bias overcloud your judgment or logic.
So if you want to continue talking about the X4640 vs DL785G6 debate, I am ok with it. But bring more data. The benchmark Sun published was rigged in Sun's favor, kind of like gang banging the F5100 with 16 HBAs, if you don't mind me bringing up an old wound.
I wouldn't have said anything more correct from "multiple of views". I said, if you have space issues but can push high power density at 10kW or more, you should consider X4640. If you don't have space issues or only can push 6kW a rack, DL785 is a better choice.
As to the rigged SAP benchmark, Sun still refused to run Linux or reduce the memory down to 128GB to take away all the advantages it wants to give itself to showcase the X4640. As long as the two software configuration and memory configurations are not the same, you cannot say X4640 is faster than DL785G6. It is simply not true.
If you want to prove that X4640 is faster, take half the memory out, and run Linux or Windows, then compare that to the score posted by DL785. Until then, you don't have any data to support your thesis that X4640 is faster than DL785.
You know what I find annoying about you? You don't seem to be the engineer who look at hard cold numbers. You let your employment bias overcloud your judgment or logic.
So if you want to continue talking about the X4640 vs DL785G6 debate, I am ok with it. But bring more data. The benchmark Sun published was rigged in Sun's favor, kind of like gang banging the F5100 with 16 HBAs, if you don't mind me bringing up an old wound.
So what, in your opinion, is superior in the Sun system to that in the hp one?
Those interesting cabling inside the system should be obvious ...
Cabling? Personally, that doesn't play a factor.
The first thing I look in a server is the form factor and price. Then the next thing I look at is whether it has any proprietary components, like proprietary disk sleds.
If the design is clearly proprietary, it whole solution gets deleted from consideration queue - because now I have clear indication that I am going to get skinned and jipped on the prices of hardware. Disks and memory is where vendors like to get you.
And that's so very, very bad. For business.
Other than that, you didn't make it clear what you considered to be better/worse, which is why I asked.
The first thing I look in a server is the form factor and price. Then the next thing I look at is whether it has any proprietary components, like proprietary disk sleds.
If the design is clearly proprietary, it whole solution gets deleted from consideration queue - because now I have clear indication that I am going to get skinned and jipped on the prices of hardware. Disks and memory is where vendors like to get you.
And that's so very, very bad. For business.
Other than that, you didn't make it clear what you considered to be better/worse, which is why I asked.
Ok, lets make some photos from SUN-Standardracks without cablecarriers, the standard rack by sun ? People who live in glass houses shouldn't throw stones.
You're right ! But since Knürr AG is the main producer of racks for Sun Microsystems, the Sun Rack and the cabling looks pretty nice. 
German quality products....
German quality products....
Jörg,
you are missing the point here: A genius likes the Chaos. There are many genius engineers at your competitors.
My office looks the same as their boxes inside!
I sometimes find things that I am looking for and not often my chair roles over my cables, so no problem at all.
Only when I would run a missioncritical application in my office, I would make sure that my cabling would be different...
Take it easy.
Frank
you are missing the point here: A genius likes the Chaos. There are many genius engineers at your competitors.
My office looks the same as their boxes inside!
I sometimes find things that I am looking for and not often my chair roles over my cables, so no problem at all.
Only when I would run a missioncritical application in my office, I would make sure that my cabling would be different...
Take it easy.
Frank
One of the most important benefits of the SUN servers:
You don't have to pay extra for an iLO2 license to get a decent iLOM.
You don't have to pay extra for an iLO2 license to get a decent iLOM.
I don't think anyone at Sun is really permitted to criticise other server manufacturers about their chassis designs. Sun has produced far too many failures in that area in the last few years.
- The PCIe cards in the T5xx0 machines aren't even close to being securely fixed, the removable support beam is total crap. I can't recall how many times our qlc or 10G cards became dislodged when plugging in a cable (which requires a little force due to the spring-loaded optical plugs), resulting in lots of funny failures and me ending up re-assembling half of the system. Also, I haven't found any screwdriver yet that does not destroy these green plastic-coated screws for the support beam and riser boards on first or second contact.. and even when screwed down tightly, the pci card holding beam isn't really stable.
- Sun can't even offer clear advice on upgrade possibilities for their servers. It's stupid to expect the customer to read through all system handbooks to find out that the T5120 can be expanded from 4 to 8 disks if he's willing to rip apart half the chassis and exchange a backplane. The parts are available, but it doesn't seem to be orderable in that config. WTF?
- The cable management arms that come with the current silver server series are crap as well, we're exchanging them by the dozens each year. The weak alu strips bend far too easily, the supporting beam in the middle rips out of its plastic clips. Please find out where IBM buys their CMAs and get the same ones, these work great.
Regarding remote management.. while the HP iLO is nicely usable, I find it rather strange that it requires a separate license. IBM RSA isn't bad either, just the remote video console tends to be a bit flaky.
Sun ILOM is.. well.. Sun ILOM. Awful from beginning to end. I wonder if the person(s) who wrote that have ever tried remotely managing a data center with it. The text console becomes usable after switching back to old ALOM mode (a feature that I hope they won't ever remove...), the web interface is useless for anything except remote powercycling.. but I could write several pages of ranting about that. Sun should really copy some features from IBM or HP there.
- The PCIe cards in the T5xx0 machines aren't even close to being securely fixed, the removable support beam is total crap. I can't recall how many times our qlc or 10G cards became dislodged when plugging in a cable (which requires a little force due to the spring-loaded optical plugs), resulting in lots of funny failures and me ending up re-assembling half of the system. Also, I haven't found any screwdriver yet that does not destroy these green plastic-coated screws for the support beam and riser boards on first or second contact.. and even when screwed down tightly, the pci card holding beam isn't really stable.
- Sun can't even offer clear advice on upgrade possibilities for their servers. It's stupid to expect the customer to read through all system handbooks to find out that the T5120 can be expanded from 4 to 8 disks if he's willing to rip apart half the chassis and exchange a backplane. The parts are available, but it doesn't seem to be orderable in that config. WTF?
- The cable management arms that come with the current silver server series are crap as well, we're exchanging them by the dozens each year. The weak alu strips bend far too easily, the supporting beam in the middle rips out of its plastic clips. Please find out where IBM buys their CMAs and get the same ones, these work great.
Regarding remote management.. while the HP iLO is nicely usable, I find it rather strange that it requires a separate license. IBM RSA isn't bad either, just the remote video console tends to be a bit flaky.
Sun ILOM is.. well.. Sun ILOM. Awful from beginning to end. I wonder if the person(s) who wrote that have ever tried remotely managing a data center with it. The text console becomes usable after switching back to old ALOM mode (a feature that I hope they won't ever remove...), the web interface is useless for anything except remote powercycling.. but I could write several pages of ranting about that. Sun should really copy some features from IBM or HP there.
Then you (or whoever integrates your servers) must be doing something really wrong. We've got scores of T5xx0s and, though there is more play in the PCIe slots then I'd like, we've never had a problem like what you describe.
My "problem" is that Sun has had much better concepts in earlier machines (take e.g. the V240 with screwed multi-slot holding brackets, or the V880 with per-card holding clips [ok, these would have been even better if they were metal instead of plastic]), but they all offered easy access and still a good hold for the cards even under mechanical stress from outside. The T5xx0 don't offer that robustness anymore.
The case interior may be as clean as it can, but I'd rather have two or three dangling cables and know my cards securely fixed instead. HP and IBM can still do that, so Sun should be able, too.
The case interior may be as clean as it can, but I'd rather have two or three dangling cables and know my cards securely fixed instead. HP and IBM can still do that, so Sun should be able, too.
"Sun ILOM is.. well.. Sun ILOM. Awful from beginning to end. I wonder if the person(s) who wrote that have ever tried remotely managing a data center with it. The text console becomes usable after switching back to old ALOM mode (a feature that I hope they won't ever remove...)"
What he said, I'm with the poster on that one!
ALOM rocks, ILOM is... garbage?
Please Joerg, if you know who makes these decisions, don't let us hang out here in the real world! Please lobby to have ALOM kept and the ILOM junk ditched.
ALOM is one of the major benefits that differentiates Sun gear from the competition... replace it with ILOM and Sun will have shot themselves in the foot. Again.
Please help us and lobby for ALOM if you can! Pretty please with sugar on top!
What he said, I'm with the poster on that one!
ALOM rocks, ILOM is... garbage?
Please Joerg, if you know who makes these decisions, don't let us hang out here in the real world! Please lobby to have ALOM kept and the ILOM junk ditched.
ALOM is one of the major benefits that differentiates Sun gear from the competition... replace it with ILOM and Sun will have shot themselves in the foot. Again.
Please help us and lobby for ALOM if you can! Pretty please with sugar on top!
+1
The LKSF book
The book with the consolidated Less known Solaris Tutorials is available for download here
Web 2.0
Contact
Networking
xing.com
LinkedIn
Facebook
My photos
- mail: joerg@c0t0d0s0.org
- icq: 144000169
- Me, myself and I
- openBC
- My del.icio.us
- c0t0d0s0@twitter
- c0t0d0s0@last.fm
- wiki@c0t0d0s0
- Amazon Wishlist
Networking
xing.com
My photos
Buttons
This work is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 3.0 Germany License
Blog Administration
Donate
Okay, okay ... as several people have asked for it ... but you know my opinion.
Tracked: Dec 12, 16:54