On Fri, Nov 11, 2005 at 03:15:48PM +0100, Oliver Neumann [New Identity AG]
> >> I've got a machine that should serve 2 XenUs as mysql cluster nodes with
> >> realy big memory consumption. As mysql cluster will need approx. 4GB memory
> >> on each XenU, I wondered if I could use hard disc space as memory for the
> >> XenU that appears as real physical memory within the VM. Swap is no
> >> alternative as mysql cluster nodes check the physical memory and if that is
> >> not enough, they will not work.
> >> How can I do that?
> >Invest a few extra ??? and put 8Gigs of ram in that server?
> >Reconfigure the mysql instances to run with just 2GB?
> >Seriously, mysql wants physical ram there for a reason, if that memory is
> >swapped, it would be faster for mysql to just read the blocks from its db
> No, with that amount of data in our databases we already calculated, that we
> need at least 4GB only for each cluster node, so we woul dhave to buy at least
> 8GB of ECC RAM which isn't really cheap. The 'problem' is that we only need to
> test an application on a cluster, after that we don't need that big amount of
> memory in that machine so the money is somehow wasted. As for the
> performance: That
> is irrelevant, the cluster does not need to be fast (it actually can be damn
I'm still really confused here. How damn slow can it be? If you have an
application that thinks it is in full memory (and needs full memory), and it
isn't, then you'll probably see in the neighborhood of a 1000 x slow down
(or worse). So an operation that normally takes 1 second, would instead 20
If we take that 1000x figure, and assume that a burn in test for the cluster
would take 1 hr (pretty small amount) under real ram, then the math states
that it would take 42 days to complete under emulated ram!
That seems a really bad way to test the functionality of the cluster, unless
you have 4 years you can wait for the test run to complete.
> Unfortunately there seems to be no chance to save the money using Xen with
> harddisc emulated physical RAM, right?
Why can't you configure the Guest OS to have the swap file? If swapping at
the Guest Domains doesn't work, there is no way that swapping at the
hypervisor level will work. The user process shouldn't really care how much
physical ram is there, just if the OS can give it more memory.
I think you need to rethink your numbers and what is expensive and what is
not. Tests that run for the better portion of a year waiting for results
seem pretty expensive as well, especially as it probably would take a couple
of attempts to get it working correctly. :)
Sean Dague Mid-Hudson Valley
sean at dague dot net Linux Users Group
There is no silver bullet. Plus, werewolves make better neighbors
than zombies, and they tend to keep the vampire population down.
Description: PGP signature
Xen-users mailing list