On Sun, Mar 29, 2009 at 6:45 PM, Gonzalo Tornaria <torna...@math.utexas.edu> wrote: > > On Sun, Mar 29, 2009 at 8:59 PM, William Stein <wst...@gmail.com> wrote: >>>> What is virtio? >> 10s of times faster than what? When I transfer a single large file >> over ssh between boxen.math.washington.edu (my vmware server host) and >> a guest image, the speed is 40MB/s on average. That's basically being >> limited by the speed of the virtual disk. > > What I meant is that virtio in KVM is 10s times faster than using > regular network card emulation also in KVM. The latter (which is the > default) means KVM in the host emulates a real network card, which the > guest sees as a standard network card. > > Using virtio networking I get 14MB/s transfer using ssh between host > and guest. If I don't use virtio, the transfer is less than 1MB/s > (don't remember the exact number, but it was quite lame). I haven't > tried vmware. It may be possible that vmware is faster than that even > in my hardware. Your hardware is definitely nicer than mine, so the > comparision is not fair (also, I think my hardware is somehow > misconfigured, but that's a whole different issue). > > I don't know how to fairly benchmark the virtual disk drive (neither > do I know how to benchmark a real hard drive).
Benchmarking is complicated. For raw speed this benchmark might be useful: sudo hdparm -t /dev/sda /dev/sda: Timing buffered disk reads: 306 MB in 3.00 seconds = 101.94 MB/sec That's the builting SAS disks in sage.math. I think one should benchmark what you care about, e.g., some people care about doctesting the sage tree, so that is a good benchmark for them. >> I didn't need console access to setup vmware *server* on my server. I >> just unpacked VMware-server-2.0.0-122956.x86_64.tar.gz, ran a script, >> and stuff worked. VMware has: > > I tried that and something didn't work, I don't remember what. But I > might try again. > >> * vmware player -- requires console >> * vmware workstation -- requires console >> * vmware server -- no console; uses a webapp and browser plugin to config > > I attempted to try vmware server. The player won't let me create > virtual machines, and the workstation is not free beer. I didn't try > very hard since I really wanted to give KVM a chance. I of course understand. VMware is the only closed source software I'm OK with, and only because I tend to see it as being "basically" hardware, so I draw my line differently. Also, I run literally dozens of virtual machines at once for months on end, so need something very robust and optimized. >> Note that I'm not trying to tell you to use closed source commercial >> software (vmware); I'm just telling you what I use. > > Thanks for telling people not to use closed source math software :-) I did that for years (with Magma). I think I've made up for that by now. Do *not* use closed source mathematical software. :-) Except to see what they can do so you can add better functionality to Sage... >>> Do you think memory limits (or vm choice) may be behind the lock ups I >>> get when I do symbolic calculations? Maybe the 300M of ulimit is too >>> tight for running maxima/clisp? >> >> That's possible. 300MB is too tight. If I were to make some official >> RAM recommendation for running Sage, it would be something like >> "minimum RAM: 1GB". > > I will try with 3GB virtual memory for the guest, adding 3-6 GB of > swap, and setting the ulimit for the notebook to 1GB. That may work > better. I'll report back if things improve. > > Gonzalo > > > > -- William Stein Associate Professor of Mathematics University of Washington http://wstein.org --~--~---------~--~----~------------~-------~--~----~ To post to this group, send email to sage-devel@googlegroups.com To unsubscribe from this group, send email to sage-devel-unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URLs: http://www.sagemath.org -~----------~----~----~----~------~----~------~--~---
