On Tue, 30 Mar 2004, Michael Will wrote: > > Dan Bolser wrote: > >> Hmm strange that you are getting 2GB per process, my large job normally > >> died around the 3GB on a standard linux kernel. > > > The job is my dumb approach to memory allocation ... > > > > #!/usr/bin/perl -w > > ... > > You probably found a limitation in the perl interpreter rather than in your > operating system. As pointed out earlier, you have about a 3GB limit when > using ia32 architecture. Ahhh.... that answers a question! > For opteron systems when using 64bit linux like Suse SLES8, you could have > 4G per 32bit-process and a much larger (limited only by your physical RAM > for at least all of 2004 :-) ) amount per 64bit-process. > > > Is there some way I can find out the make / model of the machines without > > going down to the server room and looking at the label? > Not really. What you can find out is some specs for your CPUs. > > Try cat /proc/cpuinfo Cool... processor : 0 vendor_id : GenuineIntel cpu family : 15 model : 2 model name : Intel(R) Xeon(TM) CPU 2.80GHz stepping : 7 cpu MHz : 2791.842 cache size : 512 KB physical id : 0 siblings : 2 fdiv_bug : no hlt_bug : no f00f_bug : no coma_bug : no fpu : yes fpu_exception : yes cpuid level : 2 wp : yes flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss ht tm bogomips : 5570.56 processor : 1 ... physical id : 0 ... processor : 2 ... physical id : 3 ... processor : 3 ... physical id : 3 ... > > > They are dual cpu (hyperthreaded) xeon ... My colegue told me that often > > on dual machines the memory is split, one block per cpu? > Thats not true for Xeons, there both CPUS access the same block of memory > via the front-side-bus. > > On the opterons however, you do have one block of memory per CPU, but again > it does not affect the maximum amount of memory in one process because each > CPU can access all of the RAM via Hypertransport. > > To you it just looks like one piece of RAM, except for that you might see > different access speed when not using 'node interleave memory access' which > averages out the effect of part of the RAM being closer to the CPU. > Thanks for the info, Cheers, Dan. > Michael Will > > >> >> The limitation actually lies in the 32 bit architecture of Xeon. > >> >> Physical 32 bit limitation is 4GB, PAE gives us up to 64 GB. For > >> >> normal > >> >> Linux, you should have about 3GB per process and maybe can tune the > >> >> kernel to use 3.5GB per process. > >> > > >> > Hmmm..... Looks like we are only getting 2GB per process (2GB per > >> > CPU?). > >> > > >> > We are using standard linux, but with a swap space below the size of > >> > the > >> > ram (2GB). Could this be a problem? (i.e. true64 style). > >> > > >> > ulimit -a > >> > core file size (blocks, -c) 0 > >> > data seg size (kbytes, -d) unlimited > >> > file size (blocks, -f) unlimited > >> > max locked memory (kbytes, -l) unlimited > >> > max memory size (kbytes, -m) unlimited > >> > open files (-n) 1024 > >> > pipe size (512 bytes, -p) 8 > >> > stack size (kbytes, -s) 8192 > >> > cpu time (seconds, -t) unlimited > >> > max user processes (-u) 7168 > >> > virtual memory (kbytes, -v) unlimited > >> > > >> > Mem: 4124724k av, > >> > Swap: 2040244k av > >> > > >> > Thanks very much for your help, > >> > > >> > Dan. > >> > > >> > > >> >> > >> >> LAI Loong-Fong > >> >> > >> >> On Mar 30, 2004, at 2:46 PM, Dan Bolser wrote: > >> >> > >> >>> > >> >>> I have a question about Xeon and memory... It looks like I have one > >> >>> Gb > >> >>> per > >> >>> CPU, and not 4 Gb for the 4 CPU without restriction. Is this problem > >> >>> at > >> >>> the hardware level? > >> >>> > >> >>> What is the maximum amount of memory a CPU can use? > >> >>> > >> >>> I heard talk of a Tb memory machine, but it was part of a 1000 node > >> >>> cluster, so I am thinking OK 1 Gb per node. > >> >>> > >> >>> Can anyone clarify this for me? > >> >>> > >> >>> Cheers, > >> >>> Dan. > >> >>> > >> >>> On Mon, 29 Mar 2004, Philip MacMenamin wrote: > >> >>> > >> >>>> On Monday 29 March 2004 05:38 am, you wrote: > >> >>>>> hello all, > >> >>>>> > >> >>>>> i'm interested to the constructiong of a linux cluster of computer > >> >>>>> for > >> >>>>> bioinformatics purpose. but i dont have a cluu about the performans > >> >>>>> it > >> >>>>> should have. is there any one who can give any suggestion? > >> >>>>> > >> >>>>> in the hope of an answer > >> >>>>> > >> >>>> Its kind of a nebulous question. > >> >>>> > >> >>>> Basically... buy Xeon || AMD two way boxes. Be boring, look at Dell > >> >>>> or > >> >>>> penguin computing or something. Dont bother with 64 bit. Think about > >> >>>> that in > >> >>>> about 2 years time. It will irritate you in 2 months after you buy > >> >>>> it, *i > >> >>>> promise*. > >> >>>> > >> >>>> Dont buy from some indy very cheap manufacturer, cause, if the nodes > >> >>>> will > >> >>>> fail, you will be left waiting... or make them give you a decent > >> >>>> guarantee. > >> >>>> Or a big box of bits. > >> >>>> > >> >>>> Look at the price of the chip : speed ratio. This graph will have an > >> >>>> inflection point, at which more money input gives diminishing > >> >>>> returns > >> >>>> speed-wise. Buy at this inflection point. This changes all the > >> >>>> time... of > >> >>>> course. There is no point in buying the *best* out there right this > >> >>>> second. > >> >>>> Just buy a little behind it, and buy an extra box or two. > >> >>>> > >> >>>> Find out what you want to run. > >> >>>> Buy as enough memory so it doesn't thrash your disks. (Or just fill > >> >>>> it with > >> >>>> memory). > >> >>>> > >> >>>> Spending about 2 grand (USD) on them a piece (I dont know what that > >> >>>> is in > >> >>>> Lira) should buy you something decent operating at 3 gigs, with a > >> >>>> couple of > >> >>>> gigs ram. > >> >>>> > >> >>>> Then have a look at > >> >>>> http://bioinformatics.org/biobrew/ > >> >>>> www.rocksclusters.org/ > >> >>>> > >> >>>> Sorry if this is very uninspiring/boring advice... its just, you > >> >>>> want > >> >>>> to keep > >> >>>> everything as simple as you can. There will be tricky bits no matter > >> >>>> how > >> >>>> simple you make everything. > >> >>>> Good luck. > >> >>>> Philip. >