System Load

System Load

[bikrant]

Hi,
        Is it possible to know how much system resources (cpu/memory load) 
that the 
netfilter module(s) is using? We are using HTB to shape our client traffic 
and there are 4 iptables rule for each client in the mangle table. I think it 
will be helpful to gather such data and graph it using mrtg.

I'm really sorry if someone has already asked about it.
Thanks a lot.

with regards,
Rohit Nepali

RE: system load

[James Harper]

I'm only wondering from a monitoring point of view. If I have a few
physical servers it will be useful to get an accurate picture of how
hard each is working with a view to balancing the virtual servers
sensibly across them.

James

> -----Original Message-----
> From: Ian Pratt [mailto:Ian.Pratt@cl.cam.ac.uk]
> Sent: Friday, 10 September 2004 17:32
> To: James Harper
> Cc: xen-devel@lists.sourceforge.net; Ian.Pratt@cl.cam.ac.uk
> Subject: Re: [Xen-devel] system load
> 
> > Where does system load live for xen related tasks?
> > If a packet is received by dom0, then bridged into dom1, how is the
> > system load proportioned between domains for the actual transfer of
data
> > between the two domains?
> >
> > I guess what I'm asking is, is there a way to report the amount of
cpu
> > time is spent in xen itself and the corresponding system load?
> 
> Very little time will be spent in Xen itself, but what there is
> will appear to be accounted to the domain that was running at the
> time.
> 
> If you have a domain that is doing a lot of I/O, this will be
> generating work for your driver domains (usually dom0). We've
> gone to every effort to do as little work in the driver domain as
> possible, but it does take some CPU to execute the hardware
> device driver, the bridging/firewalling code, and the 'backend'
> virtual driver.
> 
> It's hard to account and apportion exactly how much time the
> driver domain spends working on behalf of each of the other
> domains. If you're worried about a domain hogging too much of
> this resource then use can use tools like Linux's 'tc' to rate
> limit the amount of IO a particularly domain is allowed to do.
> 
> > It would be nice to be able to see the load on the physical
> > server as a whole for monitoring purposes, or is it sufficient
> > to simply sum up the load on all the domains?
> 
> Xen/xend already export the information you need to sum up load
> over each CPU, and hence see the total system load.
> 
> As Keir says, the load figures reported internally within each
> domain can be confused due to pre-emption. One trick we could do
> would be to hack Linux to create a dummy process to which we
> account all time when the domain isn't running. This would enable
> the load figures to add up, but I'm pretty sure that this is not
> what many Xen users want: In a 'virtual dedicated server'
> environment the owner of the physical server doesn't want to giveq
> customers too much information about what else is going on on the
> server...
> 
> 
> Ian



-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 13. Go here: http://sf.net/ppc_contest.php

system load

[James Harper]

Where does system load live for xen related tasks?
If a packet is received by dom0, then bridged into dom1, how is the
system load proportioned between domains for the actual transfer of data
between the two domains?

I guess what I'm asking is, is there a way to report the amount of cpu
time is spent in xen itself and the corresponding system load? It would
be nice to be able to see the load on the physical server as a whole for
monitoring purposes, or is it sufficient to simply sum up the load on
all the domains?

James


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM.
Deadline: Sept. 13. Go here: http://sf.net/ppc_contest.php

Re: system load

[Keir Fraser]

> Where does system load live for xen related tasks?
> If a packet is received by dom0, then bridged into dom1, how is the
> system load proportioned between domains for the actual transfer of data
> between the two domains?
> 
> I guess what I'm asking is, is there a way to report the amount of cpu
> time is spent in xen itself and the corresponding system load? It would
> be nice to be able to see the load on the physical server as a whole for
> monitoring purposes, or is it sufficient to simply sum up the load on
> all the domains?

Very little time will be spent in Xen, as all the hard work is done
within guest context. However, the load reported by domains is
currently incorrect when they are preempted --- the domain will think
that whatever was running at the time it got preempted ran for the
entire time that in fact someone else was running. :-(

We have some ideas for how to fix this, but in fact what you really
need is for Xen to record load stats on each CPU (who ran for how much
on each one) and report that to control software. That also shouldn't
be too hard to add.

 -- Keir


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM. 
Deadline: Sept. 13. Go here: http://sf.net/ppc_contest.php

Re: system load

[Ian Pratt]

> Where does system load live for xen related tasks?
> If a packet is received by dom0, then bridged into dom1, how is the
> system load proportioned between domains for the actual transfer of data
> between the two domains?
> 
> I guess what I'm asking is, is there a way to report the amount of cpu
> time is spent in xen itself and the corresponding system load? 

Very little time will be spent in Xen itself, but what there is
will appear to be accounted to the domain that was running at the
time.

If you have a domain that is doing a lot of I/O, this will be
generating work for your driver domains (usually dom0). We've
gone to every effort to do as little work in the driver domain as
possible, but it does take some CPU to execute the hardware
device driver, the bridging/firewalling code, and the 'backend'
virtual driver.

It's hard to account and apportion exactly how much time the
driver domain spends working on behalf of each of the other
domains. If you're worried about a domain hogging too much of
this resource then use can use tools like Linux's 'tc' to rate
limit the amount of IO a particularly domain is allowed to do.

> It would be nice to be able to see the load on the physical
> server as a whole for monitoring purposes, or is it sufficient
> to simply sum up the load on all the domains?

Xen/xend already export the information you need to sum up load
over each CPU, and hence see the total system load.

As Keir says, the load figures reported internally within each
domain can be confused due to pre-emption. One trick we could do
would be to hack Linux to create a dummy process to which we
account all time when the domain isn't running. This would enable
the load figures to add up, but I'm pretty sure that this is not
what many Xen users want: In a 'virtual dedicated server'
environment the owner of the physical server doesn't want to giveq
customers too much information about what else is going on on the
server...


Ian


-------------------------------------------------------
This SF.Net email is sponsored by: YOU BE THE JUDGE. Be one of 170
Project Admins to receive an Apple iPod Mini FREE for your judgement on
who ports your project to Linux PPC the best. Sponsored by IBM. 
Deadline: Sept. 13. Go here: http://sf.net/ppc_contest.php

System Load

[Rohit]

Hi,
	Is it possible to know how much system resources (cpu/memory load) that the 
netfilter module(s) is using? We are using HTB to shape our client traffic 
and there are 4 iptables rule for each client in the mangle table. I think it 
will be helpful to gather such data and graph it using mrtg.

I'm really sorry if someone has already asked about it.
Thanks a lot.

with regards,
Rohit Nepali

Re: System Load

[Ray Leach]

[-- Attachment #1: Type: text/plain, Size: 842 bytes --]

On Fri, 2003-11-14 at 11:34, Rohit wrote:
> Hi,
> 	Is it possible to know how much system resources (cpu/memory load) that the 
> netfilter module(s) is using? We are using HTB to shape our client traffic 
> and there are 4 iptables rule for each client in the mangle table. I think it 
> will be helpful to gather such data and graph it using mrtg.
> 
Just remember, iptables is only used to mark the traffic. tc is used to
classify and shape the traffic.

> I'm really sorry if someone has already asked about it.
> Thanks a lot.
> 
> with regards,
> Rohit Nepali
-- 
--
Raymond Leach <raymondl@knowledgefactory.co.za>
Network Support Specialist
http://www.knowledgefactory.co.za
"lynx -source http://www.rchq.co.za/raymondl.asc | gpg --import"
Key fingerprint = 7209 A695 9EE0 E971 A9AD  00EE 8757 EE47 F06F FB28
--

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 198 bytes --]

Re: System Load

[bikrant]

On Monday 17 November 2003 15:01, Ray Leach wrote:
> On Fri, 2003-11-14 at 11:34, Rohit wrote:
> > Hi,
> > 	Is it possible to know how much system resources (cpu/memory load) that
> > the netfilter module(s) is using? We are using HTB to shape our client
> > traffic and there are 4 iptables rule for each client in the mangle
> > table. I think it will be helpful to gather such data and graph it using
> > mrtg.
>
> Just remember, iptables is only used to mark the traffic. tc is used to
> classify and shape the traffic.

Yeah you are right. But iptables must be using some resources(like cpu time 
etc) to mark the each and every packet. We have around 2-3 Mbps traffice 
flowing at each interface and I think that does consume some system 
resources. That is what I want to know about.

I guess it sound very foolish now :)

regds,
Rohit
>
> > I'm really sorry if someone has already asked about it.
> > Thanks a lot.
> >
> > with regards,
> > Rohit Nepali

Re: System Load

[Antony Stone]

On Monday 17 November 2003 11:54 am, bikrant@wlink.com.np wrote:

> On Monday 17 November 2003 15:01, Ray Leach wrote:
> > On Fri, 2003-11-14 at 11:34, Rohit wrote:
> > > Hi,
> > > 	Is it possible to know how much system resources (cpu/memory load)
> > > that the netfilter module(s) is using? We are using HTB to shape our
> > > client traffic and there are 4 iptables rule for each client in the
> > > mangle table. I think it will be helpful to gather such data and graph
> > > it using mrtg.
> >
> > Just remember, iptables is only used to mark the traffic. tc is used to
> > classify and shape the traffic.
>
> Yeah you are right. But iptables must be using some resources(like cpu time
> etc) to mark the each and every packet. We have around 2-3 Mbps traffice
> flowing at each interface and I think that does consume some system
> resources. That is what I want to know about.

Presumably your traffic is not continuous, 24 hours a day, so here's a 
suggestion:

Measure your system load when the traffic is at its lowest (middle of the 
night perhaps?), and at its highest, then compare the two.

My guess is that with 2-3Mbits/sec you will hardly notice a thing.   Unless 
your netfilter is running on a 486 I don't think you'll be able to measure it 
(and even then I think it would do a perfectly good job).

Antony.

-- 

Documentation is like sex:
when it's good, it's very very good;
when it's bad, it's still better than nothing.

                                                     Please reply to the list;
                                                           please don't CC me.