From: Rus F. <rg...@jv...> - 2003-09-22 18:45:27
|
Hi All, I've got a customer that when running time is getting back some weird results that if under uml he runs # time sleep 5 real 0m10.783s user 0m0.000s sys 0m0.000s # time sleep 10 real 0m21.033s user 0m0.000s sys 0m0.000s I am thinking that the host is a dual cpu box as I think UML is single CPU . Would this account for it? On the host box [rghf@xeon rghf]$ time sleep 5 real 0m5.289s user 0m0.010s sys 0m0.010s [rghf@xeon rghf]$ time sleep 10 real 0m10.009s user 0m0.000s sys 0m0.000s Any input? Thanks Rus -- w: http://www.jvds.com | Linux + FreeBSD Servers from $15/mo e: rg...@jv... | Dedicated Servers from $119/mo t: +44 7919 373537 | email: su...@jv... t: 1-888-327-6330 | email: sa...@jv... |
From: Nick Craig-W. <nc...@ax...> - 2003-09-23 08:47:45
|
On Mon, Sep 22, 2003 at 07:46:22PM +0100, Rus Foster wrote: > I've got a customer that when running time is getting back some weird > results that if under uml he runs > > # time sleep 5 > > real 0m10.783s > user 0m0.000s > sys 0m0.000s [snip] We've noticed exactly this problem too and I've been investigating it recently. We are running 2.4.20 or 21 with skas on the hosts, and 2.4.20-um6 on the clients (mostly). It seems to happen when the host is busy in our experience, but it doesn't happen to all the UMLs on a host. It is more likely to happen to the low priority (high nice) UMLs than the high priority ones. We haven't tied it down yet though... > I am thinking that the host is a dual cpu box as I think UML is > single CPU. Would this account for it? We have this happen on single CPU hosts also. I think it is something to do with the host scheduling interacting (possibly beating with) with the UML scheduling. I was wondering whether changing HZ in the UML might affect it. (BTW why was 52 HZ chosen for UML?) -- Nick Craig-Wood nc...@ax... |
From: Peter <pet...@ri...> - 2004-07-01 23:04:37
|
Resurrecting this old thread.... In a UML (2.6.5, just the one UML instance on the host server):=20 time sleep 60 real 1m6.182s user 0m0.000s sys 0m0.000s On the host server (2.6.7, just the one aforementioned UML instance): time sleep 60 real 0m59.993s user 0m0.000s sys 0m0.001s I've tried on a few VPSs and with different sleep times. The difference = seems to be that the UML instance sleeps for 1.1 times the specified = time. =20 Any ideas what's up/how to fix? Regards, Peter ----- Original Message -----=20 From: "Nick Craig-Wood" <nc...@ax...> To: "Rus Foster" <rg...@jv...> Cc: <use...@li...> Sent: Tuesday, September 23, 2003 8:47 PM Subject: Re: [uml-user] Time problem > On Mon, Sep 22, 2003 at 07:46:22PM +0100, Rus Foster wrote: > > I've got a customer that when running time is getting back some = weird > > results that if under uml he runs > >=20 > > # time sleep 5 > >=20 > > real 0m10.783s > > user 0m0.000s > > sys 0m0.000s > [snip] >=20 > We've noticed exactly this problem too and I've been investigating it > recently. >=20 > We are running 2.4.20 or 21 with skas on the hosts, and 2.4.20-um6 on > the clients (mostly). >=20 > It seems to happen when the host is busy in our experience, but it > doesn't happen to all the UMLs on a host. It is more likely to happen > to the low priority (high nice) UMLs than the high priority ones. >=20 > We haven't tied it down yet though... >=20 > > I am thinking that the host is a dual cpu box as I think UML is > > single CPU. Would this account for it? >=20 > We have this happen on single CPU hosts also. >=20 > I think it is something to do with the host scheduling interacting > (possibly beating with) with the UML scheduling. >=20 > I was wondering whether changing HZ in the UML might affect it. (BTW > why was 52 HZ chosen for UML?) >=20 > --=20 > Nick Craig-Wood > nc...@ax... > |
From: Rus F. <rg...@jv...> - 2004-07-01 23:05:39
|
On Fri, 2 Jul 2004, Peter wrote: > Resurrecting this old thread.... > > In a UML (2.6.5, just the one UML instance on the host server): > time sleep 60 > > real 1m6.182s > user 0m0.000s > sys 0m0.000s > > On the host server (2.6.7, just the one aforementioned UML instance): > time sleep 60 > > real 0m59.993s > user 0m0.000s > sys 0m0.001s > > I've tried on a few VPSs and with different sleep times. The difference seems to be that the UML instance sleeps for 1.1 times the specified time. > > Any ideas what's up/how to fix? > Does the guest kernel have RTC support compiled in? (just guess work here) Rus -- e: rg...@vp... : t: 1-888-327-6330 www.jvds.com - Root on your own box www.vpscolo.com - Your next hosting company www.cherryhosting.net - Shared Hosting With Everything On Top |
From: Nick Craig-W. <ni...@me...> - 2004-07-02 08:30:43
|
On Fri, Jul 02, 2004 at 12:06:08AM +0100, Rus Foster wrote: > On Fri, 2 Jul 2004, Peter wrote: > > > Resurrecting this old thread.... > > > > In a UML (2.6.5, just the one UML instance on the host server): > > time sleep 60 > > > > real 1m6.182s > > user 0m0.000s > > sys 0m0.000s > > > > On the host server (2.6.7, just the one aforementioned UML instance): > > time sleep 60 > > > > real 0m59.993s > > user 0m0.000s > > sys 0m0.001s > > > > I've tried on a few VPSs and with different sleep times. The difference seems to be that the UML instance sleeps for 1.1 times the specified time. > > > > Any ideas what's up/how to fix? > > > Does the guest kernel have RTC support compiled in? (just guess work here) I'm pretty sure the RTC clock support in the guest kernel fixed these problems - its certainly not a problem any more for us :- Host: time sleep 10 real 0m10.006s user 0m0.000s sys 0m0.000s Guest: time sleep 10 real 0m10.004s user 0m0.000s sys 0m0.000s -- Nick Craig-Wood Tel: 0800 195 4968 Net: ni...@me... Memset Ltd Web: http://www.memset.com |
From: BlaisorBlade <bla...@ya...> - 2004-07-03 18:24:30
|
Alle 01:04, venerd=EC 2 luglio 2004, Peter ha scritto: > Resurrecting this old thread.... Activate CONFIG_UML_REAL_TIME_CLOCK and it should fix your problem. If not,= =20 then complain again. =2D-=20 Paolo Giarrusso, aka Blaisorblade Linux registered user n. 292729 |