Date: Tue, 14 Mar 2006 19:14:54 -0300 From: JoaoBR <joao@matik.com.br> To: freebsd-amd64@freebsd.org Subject: Re: amd64 slower than i386 on identical AMD 64 system? Message-ID: <200603141914.54442.joao@matik.com.br> In-Reply-To: <20060314112625.09a3ac2c.wmoran@collaborativefusion.com> References: <200603140740.38388.joao@matik.com.br> <200603141710.12822.kono@kth.se> <20060314112625.09a3ac2c.wmoran@collaborativefusion.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tuesday 14 March 2006 13:26, Bill Moran wrote: > We've been doing some tests here (on Dell Poweredge 2850) and haven't > done extensive tweaking (have tried different -O or any other compile > flags) > that is good I think because even if it could give certain benefit in certa= in=20 cases the base system should run as is. Special configs can be necessary fo= r=20 special application but should not be necessary to get a standard setup=20 suitable for what it was designed > So far, our conclusion is that running amd64 binaries on an amd64 > kernel is slower than ia32 binaries on an ia32 kernel. We're > comparing identical 2850 hardware, both kernels built with SMP > (although there seem to be some issues related to running SMP on > amd64) > I can confirm this too SMP amd64s are having constant crashes when running >2GB and <4GB of RAM. In order not getting anything wrong I am talking about X2-SMP mono-chip-MBs this is not happening on dual-chip-MB with two separate processors. I run the same hardware as UP-amd64 and it never crashes Since this crashes are more frequent with IPI_PREEMPTION I have now some=20 servers under test running without PREEMPTION at all and appearently the=20 crashes are gone Overall the amd64-SMP kernels running on X2 processors are extermly sensiti= ve=20 to non polling NICs and are crashing often. The overall performance also is= =20 bad.=20 Soon I change this cards into polling ones, seems XL is best, I do not have= =20 crashes anymore.=20 =46unny that single 64bit AMDs are running fine with non polling NICs even = when=20 running a SMP enabled kernel. Soon I put back the X2 ... boom. > We've been using ubench and pgbench (since these will be PostgreSQL > servers) to test. We're seeing that the 64b stuff runs just a bit > slower. We're also seeing that the amd64 doesn't seem to scale up > to using more than one processor, but that's an issue under investigation > (see other thread on this list) this I can not confirm, I get SMP X2-amds with ULE and 4BSD running on both= =20 cpus, same for dual-chip-MBs But I can not say anything about PGSQL at all My servers are cache servers in first place and I have some web and mail=20 server running amd64 and the cpu scheduling seems to work well. Overall I=20 have the impression that the ULE scheduler is giving better performance on = a=20 machine with more than 2MB/s going through Jo=E3o A mensagem foi scaneada pelo sistema de e-mail e pode ser considerada segura. Service fornecido pelo Datacenter Matik https://datacenter.matik.com.br
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200603141914.54442.joao>