I think I kinda got lost, my original question was that I wanted to run a NFS server (Unix homes) for 30 workstations. Should that NFS server be penalized running in a domU ? i.e. is it recommended to run it only on a dom0 ?
On Wed, Jul 23, 2008 at 12:51 PM, Zavodsky, Daniel (GE Money) <[EMAIL PROTECTED]> wrote: > The performance impact will most probably be very similar in both cases. > It doesn't matter if you are propagating a virtual block or net device, > the mechanism of transferring data between DomU and Dom0 is basically > the same - exchanging memory pages between those domains for all kinds > of I/O. So you can use either of them depending on which is better for > you from other viewpoints (like the ease of use). > > Regards, > Daniel > > > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Marc Grimme > Sent: Wednesday, July 23, 2008 11:26 AM > To: Red Hat Enterprise Linux 5 (Tikanga) discussion mailing-list > Subject: Re: [rhelv5-list] guaranteed cpu power to VMs > > On Wednesday 23 July 2008 11:08:48 Zavodsky, Daniel (GE Money) wrote: >> What exactly do you intend to do? >> If you want to compare using NFS directly in the guest via the network > >> VS using NFS in Dom0 and propagating a simple file on it to a guest >> via the block layer, I'd strongly suggest you not do the latter. The >> performance penalty of such a setup would be considerably more than >> 20%, I'd say more like 40%. I haven't tried NFS but the penalty of >> using simple files on GFS/SAN and exporting them to a DomU is in that > range. >> In PV DomUs, network I/O performance is pretty good. > What we want to compare is not mapping any loopback device up to DomU > but we want to compare the performance of mounting a nfs-share in DomU > to using a raw blockdevices which resides on iSCSI in Dom0 and directly > map the /dev/sd? > to DomU. That second approach eliminates the IP-Stack in DomU and the > first approach uses it. > > We would prefer NFS for the ease of use but are a little confused > because of performance considerations. > > -- > Gruss / Regards, > > Marc Grimme > http://www.atix.de/ http://www.open-sharedroot.org/ > > _______________________________________________ > rhelv5-list mailing list > [email protected] > https://www.redhat.com/mailman/listinfo/rhelv5-list > > > > _______________________________________________ > rhelv5-list mailing list > [email protected] > https://www.redhat.com/mailman/listinfo/rhelv5-list > _______________________________________________ rhelv5-list mailing list [email protected] https://www.redhat.com/mailman/listinfo/rhelv5-list
