Xen users - Jun 2009 - What are your bottomneck in using Xen solution?

Hello,

Given that nowadays standard Dual Quad Core server is such popular,
with Dual GB Ethernet, bundles of high capacity of SATA disks running
as RAID 10.

What are the factors that prevent you from installing additional VM on
an existing server?


e.g.

1. CPU
2. Memory
3. Disk I/O
4. Network
5. Else?

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Sun, Jun 14, 2009 at 4:45 PM, howard chen<howachen@gmail.com>
wrote:
> Hello,
>
> Given that nowadays standard Dual Quad Core server is such popular,
> with Dual GB Ethernet, bundles of high capacity of SATA disks running
> as RAID 10.
>
> What are the factors that prevent you from installing additional VM on
> an existing server?
>
>
> e.g.
>
> 1. CPU
> 2. Memory
> 3. Disk I/O
> 4. Network
> 5. Else?
That depends on the type of workload you put on domU. For example, if
it were only webservers or dev system then most likely the bottleneck
would be memory or CPU. On the other hand if it were database or mail
server, most likely it''s disk I/O throughput.

-- 
Fajar

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

In a well engineered system, all of these factors should become a bottleneck at
the same time.

In reality, network is often a limitation for me, as I rarely deploy Xen servers
with local storage - I prefer to use iSCSI shared storage because the ability to
migrate a domain is very useful.  This means that storage is attached via 1Gbps
or multiple 1Gbps ports.

CPU, RAM, and disk are all cheap.  10Gbps ethernet is not - yet.

Thank You,
Nathan Eisenberg
Sr. Systems Administrator
Atlas Networks, LLC

-----Original Message-----
From: xen-users-bounces@lists.xensource.com
[mailto:xen-users-bounces@lists.xensource.com] On Behalf Of howard chen
Sent: Sunday, June 14, 2009 2:46 AM
To: xen-users@lists.xensource.com
Subject: [Xen-users] What are your bottomneck in using Xen solution?

Hello,

Given that nowadays standard Dual Quad Core server is such popular,
with Dual GB Ethernet, bundles of high capacity of SATA disks running
as RAID 10.

What are the factors that prevent you from installing additional VM on
an existing server?


e.g.

1. CPU
2. Memory
3. Disk I/O
4. Network
5. Else?

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users





_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hello,

On Sun, Jun 14, 2009 at 8:34 PM, Nathan
Eisenberg<nathan@atlasnetworks.us> wrote:
> In reality, network is often a limitation for me, as I rarely deploy Xen
servers with local storage - I prefer to use iSCSI shared storage because the
ability to migrate a domain is very useful.  This means that storage is attached
via 1Gbps or multiple 1Gbps ports.
>
As Xen is not able to limit disk I/O usage easily, so by using iSCSI
shared storage, do you afraid it will make the system not easy to
scale?

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Sun, 2009-06-14 at 21:13 +0800, howard chen wrote:
> As Xen is not able to limit disk I/O usage easily, so by using iSCSI
> shared storage, do you afraid it will make the system not easy to
> scale?
Some would argue that it''s really not the hypervisor''s job to
ensure
disk QoS. Personally, I''d rather leave that up to device mapper.

If you have not done so, you might want to Google around for dm_ioband,
or check out its project page here:

http://people.valinux.co.jp/~ryov/dm-ioband/

Cheers,
--Tim


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

howard chen wrote:
> What are the factors that prevent you from installing additional VM on
> an existing server?
I have a single Core 2 Quad Xeon e5420, but my motherboard has a socket 
for a second Xeon which I''ll be getting soon.

But so far all I''ve tried is running on DomU at a time.  I have only 
been running HVM DomUs - WinXP SP3 and BeOS 5 Pro.

I had a problem with the GPLPV drivers on WinXP, but will be giving them 
another try soon.

BeOS 5 Pro is *absurdly* slow.  This is very weird because on native 
hardware, it is a particularly fast and efficient operating system.  But 
in a Xen DomU, it takes over twenty minutes to boot, and clicking a menu 
takes a couple minutes for the menu to appear.

I''d like to track down what the performance problems are, but I really 
don''t know where to start.

My hope is to use Xen to run all the target operating systems for a 
cross-platform audio program I''m writing.  I''d like to be able
to run
them all simultaneously sometimes, but I''m not having much luck so far.

Mike
-- 
Michael David Crawford
mdc@prgmr.com

    prgmr.com - We Don''t Assume You Are Stupid.

       Xen-Powered Virtual Private Servers: http://prgmr.com/xen

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Tim Post wrote:
> On Sun, 2009-06-14 at 21:13 +0800, howard chen wrote:
> 
>> As Xen is not able to limit disk I/O usage easily, so by using iSCSI
>> shared storage, do you afraid it will make the system not easy to
>> scale?
> 
> Some would argue that it''s really not the hypervisor''s
job to ensure
> disk QoS. Personally, I''d rather leave that up to device mapper.
> 
> If you have not done so, you might want to Google around for dm_ioband,
> or check out its project page here:
> 
> http://people.valinux.co.jp/~ryov/dm-ioband/
> 
> Cheers,
> --Tim
Hi,

What I don''t get is that on this page, patches never apply to a
released
version of the kernel, but always to a rcX development one. The one I
tried, for kernel 2.6.26 and 2.6.30 both fail to patch (there''s some
rejects on the Makefile). Why isn''t the author publishing something
that
people can actually USE ???

Thomas

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Hi,

On Mon, 2009-06-15 at 04:13 +0800, Thomas Goirand wrote:
> Hi,
> 
> What I don''t get is that on this page, patches never apply to a
released
> version of the kernel, but always to a rcX development one. The one I
> tried, for kernel 2.6.26 and 2.6.30 both fail to patch (there''s
some
> rejects on the Makefile).
Are you sure that you applied to the specified -rc (or quilt branch from
git?)
> Why isn''t the author publishing something that
> people can actually USE ???
> 
> Thomas
As far as I know, he''s been tuning and re-basing his patches per RC
(per
merge window) for some time now. I believe I saw some patches from him
during the .30 merge that got some attention.

There are also (as far as I can remember) a bunch of patches queued for
DM which may conflict, i.e. fixing write barriers for targets that span
multiple devices).

In any event, it illustrates a clear path to disk QoS without bothering
the hypervisor.

Cheers,
--Tim



_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Tim Post wrote:
> Hi,
> 
> On Mon, 2009-06-15 at 04:13 +0800, Thomas Goirand wrote:
> 
>> Hi,
>>
>> What I don''t get is that on this page, patches never apply to
a released
>> version of the kernel, but always to a rcX development one. The one I
>> tried, for kernel 2.6.26 and 2.6.30 both fail to patch
(there''s some
>> rejects on the Makefile).
> 
> Are you sure that you applied to the specified -rc (or quilt branch from
> git?)
Of course, I didn''t. I tried to apply to standard released kernels.
Anyway, the conflicts are very small, and it is easy to resolve, but
still, the issue remains: the author could take the time to make clean
patches against the kernels that people are actually using.

Thomas

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

howard chen <howachen@gmail.com> writes:
> Given that nowadays standard Dual Quad Core server is such popular,
> with Dual GB Ethernet, bundles of high capacity of SATA disks running
> as RAID 10.
> 
> What are the factors that prevent you from installing additional VM on
> an existing server?

I usually run out of ram before people start complaining of anything else
being slow, even with a single sata mirror.  Most of my systems are dual
socket quad core opterons (the oldest are 1.8Ghz bartons, the newer 
systems are 2.2Ghz Shanghai cores.)   My standard system has 32GiB ram.  
I have one box with 64GiB ram (and that one has two mirrors)  

yeah, as someone else said, the answer is ''it depends'' -  but
having extra
ram covers a multitude of sins.  Sufficient disk cache, once a box is
''warmed
up,'' makes up for a slow disk, and slow cpu is usually more tolerable
than
hitting disk, as once you hit disk, performance falls off a cliff, even
with the 15K SAS disk that doesn''t fit in my business model.  (now,
things
change some if you can afford the fastest SSDs available, but hey,
I''m competing on price here.   Spinning disk is the only possibility...
for
now.)

This, by the way, is why I use Xen and not OpenVZ or FreeBSD Jails, 
when I was using FreeBSD jails, (at the time, I was using 10K fibre
disk)   one heavy user would flush everyone else''s disk cache, making
performance unacceptable.  

when the CPU is overloaded, the performance degradation is usually
more gradual, and more tolerable.  

I choose my hardware to maximize the amount of ram I can give for a
particular price.            

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users