Date: Thu, 20 Jul 1995 19:01:34 -0700 (PDT) From: "Rodney W. Grimes" <rgrimes@gndrsh.aac.dev.com> To: rpt@miles.sso.loral.com (Richard Toren) Cc: hackers@freebsd.org Subject: Re: ramspeed results - ?? Message-ID: <199507210201.TAA10376@gndrsh.aac.dev.com> In-Reply-To: <Pine.SUN.3.91.950720205408.3746B-100000@miles> from "Richard Toren" at Jul 20, 95 08:59:20 pm
next in thread | previous in thread | raw e-mail | index | archive | help
> > I saw that this would swap itself to death when I first inspected the > code. I asked Poul-Henning Kamp about the malloc size and received: > > >For the results to be comparable you cannot change that number. > >If you want to run it with less memory, you also need to remove the > >checksum check's later. > > > >>> IF YOU CHANGE IT: DO NOT PUBLISH YOUR NUMBERS !!! <<< Though it is true you can't easily compare numbers of modified benchmarks, so very few people report enough information that makes the data usefull anyway. There are only a few million things that effect this data that I doubt one more would be significant, especially if it was clearly marked (and since the checksum value would be wrong, well....] ram-speed is not a good benchmark, plain and simple. It was written to do one small set of test conditions and if you can't meet the conditions I suggest you don't use it as a benchmark. Instead go get something more comprehensive like lmbench. > i did mention that I had only 8MB of memory. If the checksum fails, I > don't think anything will be reported. And will the summary results > (uSec/op) values really be comparable? For a 486DX2/66, I don't know the uSec/op number off the top of my head, but the MB/sec should be in the 5 to 15 range, unless you have an ASUS PCI/I-486SP3G, then it should be 14 for writting and 31 for reading, or well, thats the values for a DX4/100, but CPU speed should not drastically effect this test for the same cpu family and same external bus clock frequency. > > I think I will modify the block size, remove the checksum checks and see > what I get. What order of magnitude seems reasonable? > > ==================================================== > Rip Toren | The bad news is that C++ is not an object-oriented | > rpt@miles.sso.loral.com | programming language. .... The good news is that | > | C++ supports object-oriented programming. | > | C++ Programming & Fundamental Concepts | > | by Anderson & Heinze | > ==================================================== > On Thu, 20 Jul 1995, Rodney W. Grimes wrote: > > > > > > > I picked up ramspeed from this list a week ago or so. Ran it last night. > > > 1> Don't know how long it took, but it was over 90 minutes. > > > > :-(. > > > > > 2> iT's 486DX66, BT SCSI2 VLB controller, 8 MB memory > > ^^^^^^^^^^^ > > > > That code as supplied requires at least 8MB of totally free > > and unused memory or you machine pages faults and swaps to > > death. This means you need a machine with at least 12MB and > > usually more like 16MB to run it as supplied. > > > > > > > > Results - > > > 49005fb0 44.464 uS/op 2.25e+04 op/sec 0.086 mb/sec > > > 8938c0df 44.845 uS/op 2.23e+04 op/sec 0.085 mb/sec > > > > > > What is this telling me. How does it compare? Is the 3rd a disk swap rate > > > or something? > > > > It is meaningless given the configuration. All 3 result values are > > the same result just expressed in 3 different ways, they all have > > very simply mathmatical relations and given any 1 of them I can > > calculate the other 2. > > > > Change: > > #define TESTSIZE (8192*1024) > > > > to something like > > #define TESTSIZE (4096*1024) > > > > And boot your system single user to run the test to maximize the > > free memory pool and to make sure that no vm page fragmentation > > has happened. > > > > > > > > -- > > Rod Grimes rgrimes@gndrsh.aac.dev.com > > Accurate Automation Company Reliable computers for FreeBSD > > > -- Rod Grimes rgrimes@gndrsh.aac.dev.com Accurate Automation Company Reliable computers for FreeBSD
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199507210201.TAA10376>