|
Home > Archive > Unix administration > January 2006 > CacheFS Linux||UNIX Web server performance gains
You are viewing an archived Text-only version of the thread.
To view this thread in it's original format and/or if you want to reply to
this thread please [click here]
| Author |
CacheFS Linux||UNIX Web server performance gains
|
|
| Jesse Charbneau 2005-12-01, 8:55 pm |
| Hi Everyone,
If this is not the appropriate newsgroup, please let me know.
I recently starting working for a company that has implemented a 3 tier
eCommerce system. This is comprised of web servers in tier 1, app
servers in teir 2, and then the databases on the backend. Prior to
this position, I worked using Solaris in a ISP environment, and we had
used cachefs and read-only nfs mounts to serve up html/php sites. The
setup at my new position has the files on local filesystems, spread
across multiple systems (read, copy new files to multple web/app
servers). I am curious about the impact of moving the systems to a
more NFS based approach using CacheFS. Would the performance be better
in a NFS/CacheFS environment, or pulling from local disks'? It seems
that the sites I managed while working for the ISP always worked very
well, and were very easily updated. The sites I now manage all seem to
work well, but with more overhead in terms of daily management, etc.
Has anyone made changes and actually seen improvements going from a
local file system to a NFS+CacheFS setup?
Something else I am curious about; say a NFS environment does work
better, does anyone have a utility to scour the website and pull all
files and basically get them loaded into the cache, or would wget
provide enough of a pull using a recursive get?
Other things to know about the new environments:
Redhat AS EL3 w/ Apache 2.0
Resin Application Server Professional 2.1
Oracle 9i (running on a Veritas Cluster).
Thanks,
Jess
| |
| Michael Vilain 2005-12-02, 5:54 pm |
| In article <1133471450.683590.9860@g43g2000cwa.googlegroups.com>,
"Jesse Charbneau" <groups@thecharbneaus.com> wrote:
> Hi Everyone,
> If this is not the appropriate newsgroup, please let me know.
>
> I recently starting working for a company that has implemented a 3 tier
> eCommerce system. This is comprised of web servers in tier 1, app
> servers in teir 2, and then the databases on the backend. Prior to
> this position, I worked using Solaris in a ISP environment, and we had
> used cachefs and read-only nfs mounts to serve up html/php sites. The
> setup at my new position has the files on local filesystems, spread
> across multiple systems (read, copy new files to multple web/app
> servers). I am curious about the impact of moving the systems to a
> more NFS based approach using CacheFS. Would the performance be better
> in a NFS/CacheFS environment, or pulling from local disks'? It seems
> that the sites I managed while working for the ISP always worked very
> well, and were very easily updated. The sites I now manage all seem to
> work well, but with more overhead in terms of daily management, etc.
>
> Has anyone made changes and actually seen improvements going from a
> local file system to a NFS+CacheFS setup?
>
> Something else I am curious about; say a NFS environment does work
> better, does anyone have a utility to scour the website and pull all
> files and basically get them loaded into the cache, or would wget
> provide enough of a pull using a recursive get?
>
> Other things to know about the new environments:
>
> Redhat AS EL3 w/ Apache 2.0
> Resin Application Server Professional 2.1
> Oracle 9i (running on a Veritas Cluster).
>
> Thanks,
> Jess
Think very hard about implementing an Linux NFS solution if you need it
to talk to anything other than another Linux system. Some versions of
the Linux kernel are just plain broken when it comes to NFS. They don't
work well with commercial implementations of NFS clients (e.g. HP/UX,
Solaris, or AIX). Switch to a commercial UNIX NFS server and Linux
client NFS and the problem goes away.
Before you start messing with what already works, test what you propose
on a test cluster and private network, then make that test cluster
public and have it banged on, then plan to move it into production. If
you proceed this way, you'll probably stay in this job a lot longer.
--
DeeDee, don't press that button! DeeDee! NO! Dee...
| |
| Chris Cox 2005-12-04, 2:48 am |
| Michael Vilain wrote:
> In article <1133471450.683590.9860@g43g2000cwa.googlegroups.com>,
> "Jesse Charbneau" <groups@thecharbneaus.com> wrote:
>
>
> Think very hard about implementing an Linux NFS solution if you need it
> to talk to anything other than another Linux system. Some versions of
> the Linux kernel are just plain broken when it comes to NFS. They don't
> work well with commercial implementations of NFS clients (e.g. HP/UX,
> Solaris, or AIX). Switch to a commercial UNIX NFS server and Linux
> client NFS and the problem goes away.
>
> Before you start messing with what already works, test what you propose
> on a test cluster and private network, then make that test cluster
> public and have it banged on, then plan to move it into production. If
> you proceed this way, you'll probably stay in this job a lot longer.
>
FUD. The assumption that commercial
Unix NFS works is just stupid. Commercial Unix has the same
ammount if not more bugs than Linux... except it has only
been of late that those vendors have found the need to
fix their problems more quickly.
We run large 10+ terabytes of Linux exported NFS in a mixed
commercial Unix enviroment with just about every version
of NFS imaginable. No problems.
If anything, the Linux boxes deal better with strange
situations, whereas (for example) the Solaris boxes
give up (and sometimes lock up) if the NFS share isn't
available for a bit (due to a network problem).
Our original NFS infrastructure ran on Solaris 7.
We switched it to SUSE Linux and our performance
increased and our costs dramatically decreased.
I find the Linux NFS plays with more Unix systems
than NFS from (for example) Solaris.
| |
| Jesse Charbneau 2005-12-09, 5:58 pm |
| Hi,
Thanks for the reply Chris. Would you say that the performance is
better with NFS than with local disks'? Not trashing NFS or anything,
as I actually prefer it, but would like some ammo when I start
suggesting (Probably down the road since I'm the new guy) a switch to
NFS for ease of management. Like I was saying, we have a dozen
eCommerce sites, each having at least two web servers, and several app
servers for all the sites, and we currently are copying the files to
each individual box, which from an ease of mgmt point of view is a
major pain.
As far as NFS on Linux being broken, I'm definitely not an expert here
as my experience is mainly with the BSDs and Solaris. Years ago I used
NFS with Linux, but was not very schooled in NFS/NIS at the time. I
have seen the infamous system locks when NFS server is unavailable
though (mainly with Slowaris), and it is no small matter to reboot a
server that needs 99.9% uptime.
Does anyone have any info regarding using CacheFS on Linux as you can
in Solaris? I am very interested in this, although our JSP apps have
some type of Caching available, but I would prefer to do this at the
system level instead of trusting the app servers to do it for me.
Thanks again everyone,
Jess
BTW Our kernel version is in the 2.4.x series (RH AS EL3).
Chris Cox wrote:
> Michael Vilain wrote:
>
> FUD. The assumption that commercial
> Unix NFS works is just stupid. Commercial Unix has the same
> ammount if not more bugs than Linux... except it has only
> been of late that those vendors have found the need to
> fix their problems more quickly.
>
> We run large 10+ terabytes of Linux exported NFS in a mixed
> commercial Unix enviroment with just about every version
> of NFS imaginable. No problems.
>
> If anything, the Linux boxes deal better with strange
> situations, whereas (for example) the Solaris boxes
> give up (and sometimes lock up) if the NFS share isn't
> available for a bit (due to a network problem).
>
> Our original NFS infrastructure ran on Solaris 7.
> We switched it to SUSE Linux and our performance
> increased and our costs dramatically decreased.
>
> I find the Linux NFS plays with more Unix systems
> than NFS from (for example) Solaris
| |
| Chris Cox 2005-12-14, 5:57 pm |
| Jesse Charbneau wrote:
> Hi,
> Thanks for the reply Chris. Would you say that the performance is
> better with NFS than with local disks'? Not trashing NFS or anything,
NFS performance is certainly not better than local disk. Even with a
gigabit infrastructure. Just not enough bandwidth.
CacheFS might be an ok choice. Not sure.
> as I actually prefer it, but would like some ammo when I start
> suggesting (Probably down the road since I'm the new guy) a switch to
> NFS for ease of management. Like I was saying, we have a dozen
> eCommerce sites, each having at least two web servers, and several app
> servers for all the sites, and we currently are copying the files to
> each individual box, which from an ease of mgmt point of view is a
> major pain.
>
> As far as NFS on Linux being broken, I'm definitely not an expert here
> as my experience is mainly with the BSDs and Solaris. Years ago I used
> NFS with Linux, but was not very schooled in NFS/NIS at the time. I
> have seen the infamous system locks when NFS server is unavailable
> though (mainly with Slowaris), and it is no small matter to reboot a
> server that needs 99.9% uptime.
>
> Does anyone have any info regarding using CacheFS on Linux as you can
> in Solaris? I am very interested in this, although our JSP apps have
> some type of Caching available, but I would prefer to do this at the
> system level instead of trusting the app servers to do it for me.
I'd go through the cachefs mail list for redhat and see if they
can answer the questions.
Google for cachefs linux...
>
> Thanks again everyone,
>
> Jess
>
> BTW Our kernel version is in the 2.4.x series (RH AS EL3).
>
> Chris Cox wrote:
>
| |
| Thomas Maier-Komor 2006-01-23, 7:52 am |
| Chris Cox wrote:
> Our original NFS infrastructure ran on Solaris 7.
> We switched it to SUSE Linux and our performance
> increased and our costs dramatically decreased.
>
Chris, don't you think it is a bit unfair to compare the eight years old
Solaris 7 with a recent Linux distribution? Have you ever done a
comparison of Solaris 10 and a current Linux distribution? Solaris 7 is
really slow and unstable compared to 10.
> I find the Linux NFS plays with more Unix systems
> than NFS from (for example) Solaris.
>
My experience is different. We had _many_ problems with Linux NFS and
never a real problem with a NFS shared served by a solaris host.
YMMV.
Tom
|
|
|
|
|