hi, i tested gigabit eth and 100 mbit ethernet cards on some systems and i noticed that bottleneck was generated between CPU and disk. Simply, you can see it yourself with helping copy/http get/write and time utility.I remember there are alot articles on the net regarding system bottlenecks. You can use google..:-) Also, cluster server doenst make your sytems more performance. It makes your systems single point of failure or fault tolerans. Hope this help.. Ilker G. Hong Hsu wrote: > Chris, Many Thanks for your help AND point out my one typo. > > Actually I am thinking to add power to my system, either using Linux's SMP > and dual Intel Pentium 4 processors with a single system bus, OR clustering > two single processor Linux machines. With help of 64-bits 1GB Network card > and CAT5e cable (I think 350 Mbps), network part doesn't seem a bottleneck. > But I not sure which approach has better performance. > > Thanks again, > -Hong > > > "Christopher P Wright " wrote: > > >>>I have general questions regarding a typical Linux system's speed and >>>wonder this is right place to ask these questions. If this is not, Could >>>someone point out which group I can post. >>> >>probably not the best place, but i know of nowhere else. >> >> >>>With following typical components on a motherboard: >>> 512 MB 10K RPM DRAM, >>> Intel 850 chipset with 64-bits Data bus width and 400MHz Data Rate, >>> 32-bit/64-bit PCI 2.10 bus (33MHz/66MHz) >>> 20 GB Hard Drive >>> >>i think the 10k rpm goes with the harddrive, as dram doesn't rotate =) >> >> >>> Does Intel 850's 400MHz data rate fully used or not on 66MHz bus >>>speed? As speed of these components are measured by rates, I am >>>wondering how one can evaluate the system's speed roughly in terms of >>>using MB/sec so that one can see potential bottleneck or trend of >>>improvement. On Windows, using PCMark2002 benchmark software from >>>MadOnion.com, one can see 20-70 MB/sec on HD, 700 - 1,400 MB/sec on >>>DRAM. Is there a way one can measure system bus actual speed under >>>Linux or benchmark for DRAM, HD? >>> >>the 400Mhz is the dram clock. 66Mhz is standard pci (as is 33 sometimes). >>obviously, the 400Mhz would not be used fully from a 66Mhz feed. this is >>typical. (memory is faster than pci cards, etc). >> >>to test HD performance one can use 'hdparm -T -t'. that benchmarks hd >>speeds, and buffered reads (sort of memory bandwidth maybe?). im not sure >>of a tool to measure dram bandwidth directly, but i'm sure they exist >>somewhere. be sure to enable dma on the harddrive before you benchmark >>it, or the speeds will be worse. ( 'hdparm -d1 [device]' ) >> >> >>>In addition, if I like to add a 1GB Network Interface Card for >>>clustering two same machines, should I add a 32-bit NIC card or 64-bit >>>NIC in terms of performance, and why? >>> >>64bit would generally be able to transfer data to/from the system twice >>as fast as a 32bit card, simply because it transfers twice as many bits >>per cycle. i think 64 bit may be clocked higher (the 66mhz) than the 32 >>bit (33 mhz???) but i'm foggy on pci specs in that regard, someone else >>probably knows far more in depth. >> >>ttyl >>chris >> >>-- >>"In other words, I'm lost, don't know where we are, >>where we're going, or even if we're going anywhere, >>and don't have control anyway. Otherwise everything's fine." >> > > -- > <Linux kernel:>< As a computer I find your faith in technology amusing. > > > -- > Kernelnewbies: Help each other learn about the Linux kernel. > Archive: http://mail.nl.linux.org/kernelnewbies/ > FAQ: http://kernelnewbies.org/faq/ > > -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ FAQ: http://kernelnewbies.org/faq/
