Message boards :
Number crunching :
Sparc Vs. Intel
Message board moderation
Author | Message |
---|---|
Wanabeelee Send message Joined: 30 Jun 02 Posts: 12 Credit: 1,528,919 RAC: 0 |
I dont claim to be an expert on Sun servers. But everything that I have seen show sparc cpu's just not holding any water to intels Zeon's or amd's for that matter. Why have 4 deleter's running on one server with the 4 validaters? Would it not be better to have 2 validaters running on a duel intel and pass info from one box to another? Most of the bench marks I have seen show the 400's in sun's e3500 = about zeons 800mhz in fp and int. But really the question is why have 8 things running on a duel cpu server? |
[B@H] Ray Send message Joined: 1 Sep 00 Posts: 485 Credit: 45,275 RAC: 0 |
I dont claim to be an expert on Sun servers. But everything that I have seen show sparc cpu's just not holding any water to intels Zeon's or amd's for that matter. Why have 4 deleter's running on one server with the 4 validaters? Would it not be better to have 2 validaters running on a duel intel and pass info from one box to another? Most of the bench marks I have seen show the 400's in sun's e3500 = about zeons 800mhz in fp and int. But really the question is why have 8 things running on a duel cpu server? The processors are not the problem, it is the directory structure and number of files. If you are saying that because it the number WFV going up, here if what I posted to another thread: I think that is more from which files are comming in. I have 17 WTV files, but when checking them I find that only 1 or 2 results have been turned in. Most of these are old, processed prior to outage, a few waiting for a copy to be reissued. And 5 are new whare I am the only one to turn that unit in. This should work out as more get there old files turned in. With the Golf coast of the US not having power now more may have to wait for the last to go past the deadline & be reissued to another computer also. Many factors besides the disk and directory problems, only time will tell at this point. Ray Pizza@Home Rays Place Rays place Forums |
Wanabeelee Send message Joined: 30 Jun 02 Posts: 12 Credit: 1,528,919 RAC: 0 |
The processors are not the problem, it is the directory structure and number of files. I am saying it cause I have seen them turn off some services to speed up others. When asking a disk to access 8 diff places at a time you are going to get delays. Working in IT is basically long periods of boredom broken up by sheer moments of terror. |
vescudero Send message Joined: 10 Jul 04 Posts: 7 Credit: 967,048 RAC: 0 |
I dont claim to be an expert on Sun servers. But everything that I have seen show sparc cpu's just not holding any water to intels Zeon's or amd's for that matter. Why have 4 deleter's running on one server with the 4 validaters? Would it not be better to have 2 validaters running on a duel intel and pass info from one box to another? Most of the bench marks I have seen show the 400's in sun's e3500 = about zeons 800mhz in fp and int. But really the question is why have 8 things running on a duel cpu server? Those delays seems to be more I/O related, so faster cpus won't help to speed up the process. Apart from this, I agree that those old SPARC cpus are not very fast if you compare only one thread running on SPARC against one thread on Intel or AMD. Nevertheless, you should be aware a) Those CPUS are probably 5 years old or even more, so it would be more accurate it you compare it against new UltraSPARC IV+ cores. b) MIPS and MFLOPS are only one kind of measure for cpu's, but they don't take into account cache latency, effectiveness of the processor to swap registers in only a few cpu cycles, branch prediction hit rates, ... etc. c) Nowadays x86 architecture is restricted by some scalability issues, while SPARC scales almost linearly on Solaris up to more than one hundred cpus. For example on x86 or x64 cpus you will notice that 32 generic processes running at the same time on one cpu will perform worst in terms of work, that only one process running in one cpu at full speed. On SPARC you will find out that each process will do about 1/32 of the work and the total work done will be the same as done by only one process running without competition. On Intel you will find out that such a huge amount of processes (only 32!) will probably make your computer act as hung and once the processes are finished, the total work done by all of them will not be more than 2/3 of the work done by only one process running on one cpu. |
Scottatron Send message Joined: 15 Jul 03 Posts: 94 Credit: 220,389 RAC: 0 |
I dont claim to be an expert on Sun servers. But everything that I have seen show sparc cpu's just not holding any water to intels Zeon's or amd's for that matter. Why have 4 deleter's running on one server with the 4 validaters? Would it not be better to have 2 validaters running on a duel intel and pass info from one box to another? Most of the bench marks I have seen show the 400's in sun's e3500 = about zeons 800mhz in fp and int. But really the question is why have 8 things running on a duel cpu server? lol, I would NEVER have guessed that someone named Sun-Microsystems would favour Sun products :D |
Wanabeelee Send message Joined: 30 Jun 02 Posts: 12 Credit: 1,528,919 RAC: 0 |
Thanks for the info on the sparc cpu. Sparc cpu's being able to handle more processes sounds fair. But if the problem is in data access times, how would having 4 possesses requesting 4 diff files help in I/O times? I know I keep my networks user files on a sep drive array than I do the exchange box cause of the delay in accessing stores when you have to many request for diff data at the same time. Having too many data request on any of my raid 5 arrays seems to have the same effect. Would it not be better to have single processes making more data request in sequence than having 4 making request for data at the same time? I know my nas boxes try to handle multi request at the same time. lol, I would NEVER have guessed that someone named Sun-Microsystems would favour Sun products :D But you would think that they might know a thing or two about sun products. There must be a reason why Web MD has over 20 sun servers in the nashville branch. Working in IT is basically long periods of boredom broken up by sheer moments of terror. |
Uioped1 Send message Joined: 17 Sep 03 Posts: 50 Credit: 1,179,926 RAC: 0 |
Would it not be better to have single processes making more data request in sequence than having 4 making request for data at the same time? Actually you've got it exactly backwards. Even on a 400 mhz cpu, computation and memory access will be _much_ faster than disk IO, even on local drives. So, if you have one process emitting IO requests very fast, your results will be more time spent blocked on those IO requests. Even if had some fancy non-blocking scheme that allowed the process to continue working until it needed the results of that IO, you would find that you'll be better off with multiple requests from different processes. Furthermore, raid is a technology that only delivers performance increases if you have parallel data requests, by splitting volumes across multiple drives and optimizing the request order. |
Sir Ulli Send message Joined: 21 Oct 99 Posts: 2246 Credit: 6,136,250 RAC: 0 |
what do you think about this http://www.sun.com/servers/highend/sunfire_e25k/index.xml okay Berkely has not the newest Hardware, but they doing their job, i think.. Greetings from Germany NRW Ulli |
SunMicrosystemsLLG Send message Joined: 4 Jul 05 Posts: 102 Credit: 1,360,617 RAC: 0 |
what do you think about this I think you are dreaming... |
ThePhantom86 Send message Joined: 30 Jan 02 Posts: 268 Credit: 1,970,082 RAC: 0 |
Where I work, we don't have a E25K but we do have a E10K and F12K. Unless someone wants to donate one to Berkeley, they ain't gonna get one. Power consumption alone would hurt Berkeley. |
©2024 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.