linux and high volume web sites

linux and high volume web sites

[valery]

I have a high volume web site under linux :
kernel is 2.2.17
hardware is 5 bi-PIII 700Mhz / 512Mb, eepro100
all server are diskless (nfs on an netapp filer) except for tmp and swap

dispatch is done by the Resonate product

web server is apache+php (something like 400 processes), database
backend is a mysql on the same hardware


in high volume from time to time machines are "freezing" then after a
few seconds they "reappear" and response timne is 


how can I investigate all these problems ?

Re: linux and high volume web sites

[David Lang]

watch the resonate heartbeat and see if it is getting lost in the network
traffic (the resonate logs will show missing heartbeat packets). think
seriously of setting the resonate stuff to run at a higher priority so
that it doesn't get behind.

depending on how high your network traffic is seriously look at putting in
a second nic and switch to move the NFS traffic off the network that has
the internet traffic and hearbeat.

I had the same problem with central dispatch a couple years ago when first
implementing it. the exact details of the problem that I ran into should
have been fixed by now (mostly having to do with large number of virtual
IP addresses) but the symptoms were the same.

David Lang


 On
Sat, 28 Apr 2001, valery wrote:

> Date: Sat, 28 Apr 2001 11:01:19 +0200
> From: valery <valery.brasseur@atosorigin.com>
> To: linux kernel <linux-kernel@vger.kernel.org>
> Subject: linux and high volume web sites
>
> I have a high volume web site under linux :
> kernel is 2.2.17
> hardware is 5 bi-PIII 700Mhz / 512Mb, eepro100
> all server are diskless (nfs on an netapp filer) except for tmp and swap
>
> dispatch is done by the Resonate product
>
> web server is apache+php (something like 400 processes), database
> backend is a mysql on the same hardware
>
>
> in high volume from time to time machines are "freezing" then after a
> few seconds they "reappear" and response timne is
>
>
> how can I investigate all these problems ?
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>

Re: linux and high volume web sites

[Tim Moore]

David Lang wrote:
> 
> watch the resonate heartbeat and see if it is getting lost in the network
> traffic (the resonate logs will show missing heartbeat packets). think
> seriously of setting the resonate stuff to run at a higher priority so
> that it doesn't get behind.
> 
> depending on how high your network traffic is seriously look at putting in
> a second nic and switch to move the NFS traffic off the network that has
> the internet traffic and hearbeat.
> 
> I had the same problem with central dispatch a couple years ago when first
> implementing it. the exact details of the problem that I ran into should
> have been fixed by now (mostly having to do with large number of virtual
> IP addresses) but the symptoms were the same.

In addition to the above make sure there's enough bandwidth to the filer
(eg- good switches, multiple ethernets).

Consider moving to 2.2.19.  Significant VM changes after 2.2.19pre3 which
could account for the freezes.

rgds,
tim.

> > I have a high volume web site under linux :
> > kernel is 2.2.17
> > hardware is 5 bi-PIII 700Mhz / 512Mb, eepro100
> > all server are diskless (nfs on an netapp filer) except for tmp and swap
> >
> > dispatch is done by the Resonate product
> >
> > web server is apache+php (something like 400 processes), database
> > backend is a mysql on the same hardware
> >
> > in high volume from time to time machines are "freezing" then after a
> > few seconds they "reappear" and response timne is
> >
> >
> > how can I investigate all these problems ?

--

Re: linux and high volume web sites

[Sean Hunter]

Also make sure you aren't suffering database lock contention from Mysql.  This
causes very fast context switching on the database server, and is typically
unable to do useful work even though its load avg is not high.  "vmstat" is
useful here.

Sean

On Sat, Apr 28, 2001 at 01:55:01PM -0700, Tim Moore wrote:
> David Lang wrote:
> > 
> > watch the resonate heartbeat and see if it is getting lost in the network
> > traffic (the resonate logs will show missing heartbeat packets). think
> > seriously of setting the resonate stuff to run at a higher priority so
> > that it doesn't get behind.
> > 
> > depending on how high your network traffic is seriously look at putting in
> > a second nic and switch to move the NFS traffic off the network that has
> > the internet traffic and hearbeat.
> > 
> > I had the same problem with central dispatch a couple years ago when first
> > implementing it. the exact details of the problem that I ran into should
> > have been fixed by now (mostly having to do with large number of virtual
> > IP addresses) but the symptoms were the same.
> 
> In addition to the above make sure there's enough bandwidth to the filer
> (eg- good switches, multiple ethernets).
> 
> Consider moving to 2.2.19.  Significant VM changes after 2.2.19pre3 which
> could account for the freezes.
> 
> rgds,
> tim.
> 
> > > I have a high volume web site under linux :
> > > kernel is 2.2.17
> > > hardware is 5 bi-PIII 700Mhz / 512Mb, eepro100
> > > all server are diskless (nfs on an netapp filer) except for tmp and swap
> > >
> > > dispatch is done by the Resonate product
> > >
> > > web server is apache+php (something like 400 processes), database
> > > backend is a mysql on the same hardware
> > >
> > > in high volume from time to time machines are "freezing" then after a
> > > few seconds they "reappear" and response timne is
> > >
> > >
> > > how can I investigate all these problems ?
> 
> --
> -
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>