Hi,
I have a question about the benchmarks published on the Xen website, mainly
how UML was benchmarked. My point is that the details aren''t described
anywhere, and there are several tricks which dramatically improve UML
performance, mainly IO-Latency and scheduling.
So, here comes:
- what was the data, such as rootfs, in reality? was it a raw partition or an
ubd or hostfs (or whatever)?
- was the UML memory saved on tmpfs?
- was swap used inside the guest, again was it raw or ubd or whatever?
Anyway, Xen benchmarks look very impressive.
Bye,
Peter Surda (Shurdeek) <shurdeek@routehat.org>, ICQ 10236103,
+436505122023
--
NT, now approaching 23x6 availability.
-------------------------------------------------------
This SF. Net email is sponsored by: GoToMyPC
GoToMyPC is the fast, easy and secure way to access your computer from
any Web browser or wireless device. Click here to Try it Free!
https://www.gotomypc.com/tr/OSDN/AW/Q4_2003/t/g22lp?Target=mm/g22lp.tmpl
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/xen-devel
On Sun, 2003-11-16 at 09:22, Peter Surda wrote:> Hi, > > I have a question about the benchmarks published on the Xen website, mainly > how UML was benchmarked. My point is that the details aren''t described > anywhere, and there are several tricks which dramatically improve UML > performance, mainly IO-Latency and scheduling. > > So, here comes: > - what was the data, such as rootfs, in reality? was it a raw partition or an > ubd or hostfs (or whatever)? > - was the UML memory saved on tmpfs? > - was swap used inside the guest, again was it raw or ubd or whatever? >I haven''t personally used UML, but it''s always good to keep options open. Could you provide which you perceive to have the best performance for each of these that you ask (ie, ubd or hostfs). I''d like to test this myself and if you know anything that would improve UML performance i''d like to include it. It doesn''t look like UML is anywhere near close to providing the QoS of Xen, but if the server isn''t expected to overreach it''s limits then perhaps that isn''t a huge problem.> Anyway, Xen benchmarks look very impressive. > > Bye, > > Peter Surda (Shurdeek) <shurdeek@routehat.org>, ICQ 10236103, +436505122023-- Matt Ayres <matta@tektonic.net> TekTonic ------------------------------------------------------- This SF. Net email is sponsored by: GoToMyPC GoToMyPC is the fast, easy and secure way to access your computer from any Web browser or wireless device. Click here to Try it Free! https://www.gotomypc.com/tr/OSDN/AW/Q4_2003/t/g22lp?Target=mm/g22lp.tmpl _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel
> Hi, > > I have a question about the benchmarks published on the Xen website, mainly > how UML was benchmarked. My point is that the details aren''t described > anywhere, and there are several tricks which dramatically improve UML > performance, mainly IO-Latency and scheduling.The SOSP paper contains more details. The UML was running with the SKA host patches.> So, here comes: > - what was the data, such as rootfs, in reality? was it a raw partition or an > ubd or hostfs (or whatever)?The root file system and all data were stored in a real (raw) disk partition containing ext3.> - was the UML memory saved on tmpfs?No special measures were taken to control where the UML memory went, as I''m afraid the UML howtos didn''t indicate that this was important. If the default is to send it to /tmp (?), that would have been the host''s root ext3. If I get a chance, I''ll rerun the UML results with a tmpfs.> - was swap used inside the guest, again was it raw or ubd or whatever?None of the experiments caused the guests to swap, so I believe swap was left unconfigured (I haven''t got access to the machine right now to check) Best, Ian> Anyway, Xen benchmarks look very impressive. > > Bye, > > Peter Surda (Shurdeek) <shurdeek@routehat.org>, ICQ 10236103, +436505122023 > > -- > NT, now approaching 23x6 availability. > > > ------------------------------------------------------- > This SF. Net email is sponsored by: GoToMyPC > GoToMyPC is the fast, easy and secure way to access your computer from > any Web browser or wireless device. Click here to Try it Free! > https://www.gotomypc.com/tr/OSDN/AW/Q4_2003/t/g22lp?Target=mm/g22lp.tmpl > _______________________________________________ > Xen-devel mailing list > Xen-devel@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/xen-devel------------------------------------------------------- This SF. Net email is sponsored by: GoToMyPC GoToMyPC is the fast, easy and secure way to access your computer from any Web browser or wireless device. Click here to Try it Free! https://www.gotomypc.com/tr/OSDN/AW/Q4_2003/t/g22lp?Target=mm/g22lp.tmpl _______________________________________________ Xen-devel mailing list Xen-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/xen-devel