A few days ago someone was asking whether 100MHz pentium was enough to do what looked like a small routing job. My guess at the time was that much less was needed but I didn''t have any data. Now I do. I have a 486 acting as a firewall on a .5Mb DSL line. For what it''s worth, dmesg says: Calibrating delay loop... 16.38 BogoMIPS Memory: 14572k/16640k I just scp''d about 15MB over that connection (took about 7-8 min) and watched the load average while it was happening. It converged to zero.
On Wed, Jul 11, 2001 at 12:00:26PM -0700, Don Cohen wrote:> > A few days ago someone was asking whether 100MHz pentium was enough to > do what looked like a small routing job. My guess at the time was > that much less was needed but I didn''t have any data. > Now I do. I have a 486 acting as a firewall on a .5Mb DSL line. > For what it''s worth, dmesg says: > Calibrating delay loop... 16.38 BogoMIPS > Memory: 14572k/16640k > I just scp''d about 15MB over that connection (took about 7-8 min) and > watched the load average while it was happening. It converged to zero. >I don''t believe that routing load is factored into this load average. To get a measurement, you''d need some kernel profiling tools (cant think of any now though). Mike
Michael T. Babcock
2001-Jul-11 19:22 UTC
Re: On the question of how big a machine you need
Mike Fedyk wrote:>I don''t believe that routing load is factored into this load average. >To get a measurement, you''d need some kernel profiling tools (cant >think of any now though). >Although having the kernel CPU times be accurate, etc., I think its more important to know how much overall bandwidth was achieved with a 486 as a router, with and without QoS or other rules on. If the speed doesn''t get slower (except where limiting is enforced, of course), then its a big enough machine (and we can all buy up a bunch of old 486 stock to build Linux-based router appliances). :-) -- Michael T. Babcock CTO, FibreSpeed
On Wed, Jul 11, 2001 at 03:22:24AM -0400, Michael T. Babcock wrote:> >I don''t believe that routing load is factored into this load average. > >To get a measurement, you''d need some kernel profiling tools (cant > >think of any now though). > > > Although having the kernel CPU times be accurate, etc., I think its > more important to know how much overall bandwidth was achieved with a > 486 as a router, with and without QoS or other rules on. If the speed > doesn''t get slower (except where limiting is enforced, of course), then > its a big enough machine (and we can all buy up a bunch of old 486 stock > to build Linux-based router appliances).in terms of raw speed, the cpu chip doesn''t make all that much difference. a cisco 2501, for instance, is a 20Mhz 68030 (think Mac SE/30 here). a modern floppy drive controller can spew bits faster than a T-1. of more importance is probably what else is going on (filtering, dynamic routing table maintenance, packet assembly/reassembly/inspection, multiple links, fast ethernet, etc), and bus-related bottlenecks. i think either linux or freebsd, properly stripped down, could handle a T-1/ethernet combo, with modest routing rules, with ease on a 486 (just make sure you give it lots of memory, like at least 32MB). also, an efficient set of wan/lan cards will make a big difference. ref: www.linuxrouter.org (a disk-less, fan-less single-floppy linux "distribution" for "instant router" functionality). -- Henry Yen Aegis Information Systems, Inc. Senior Systems Programmer Hicksville, New York
Arthur van Leeuwen
2001-Jul-12 07:42 UTC
Re: On the question of how big a machine you need
On Wed, 11 Jul 2001, Don Cohen wrote:> I just scp''d about 15MB over that connection (took about 7-8 min) and > watched the load average while it was happening. It converged to zero.It would. The routing is done in kernelspace and therefore not counted against the load average. However, given the relative simplicity of dealing with network traffic at speeds lower than (say) 100 Mbit/s, any system with a PCI bus and a processor that has a clock multiplier on the memory-bus clock should be able to quite easily route at line rate. Thus, anything over a PII-233 should pose zero problems whatsoever, whatever you want to do. Doei, Arthur. (Ofcourse, I''m not talking GigEther here... that''s an entirely different ballpark. Standard PCI can''t even really keep up with *that*) -- /\ / | arthurvl@sci.kun.nl | Work like you don''t need the money /__\ / | A friend is someone with whom | Love like you have never been hurt / \/__ | you can dare to be yourself | Dance like there''s nobody watching