Re: [Xen-API] XCP 1.1 Intellicache issues

[Lars Seeliger <meesterlars@xxxxxxxxx>]

Hey, Andrei.

Thanks for the info; it's clear IntelliCache is not intended for my usage model (simple I/O buffer for NFS storage). We were hoping to implement this feature on all XCP installations to boost general VM performance, but we'll have to wait until a more robust implementation of IntelliCache is released.


Until then we're going to use Nexentastor for shared storage, where it's possible to enable a read cache at the file-system level, which is probably better, anyway. Just FYI, perhaps you guys should have a look at bcache as a way to speed up XCP:


http://bcache.evilpiepirate.org/

It's not production ready, but the source code is there and you lot seem to know what you're doing :p

Is there a plan to have IntelliCache optimised for general VM usage, i.e. not specific to XenDesktop? The potential of the feature is amazing - my corruption experience aside, when using a fast cache device performance is staggering.

On 28 October 2011 13:15, Andrei Lifchits <Andrei.Lifchits@xxxxxxxxxx> wrote:

Hi Lars,

 

Yes, unfortunately, this is a bug: we should not allow the snapshot operations when IntelliCache is turned on. It was a conscious decision to keep things simple, as the only user of IntelliCache at the moment is XenDesktop.

 

Just in case, below are the instructions for the XenDesktop use case.

 

To create user desktop VMs with IntelliCache optimisations, the following steps must be taken:

 

1. choose a VM or snapshot to deploy

2. call "VM.copy" to create a gold master VM

3. set the optimisation-related fields on the gold master VM

4. call "VM.clone" lots of times to create the user desktops with the right caching properties

 

Cheers,

Andrei

 

 

 

From: Lars Seeliger <meesterlars@xxxxxxxxx>

Subject: [Xen-API] XCP 1.1 Intellicache issues

Date: 14 October 2011 10:01:50 GMT+01:00

To: "xen-api@xxxxxxxxxxxxxxxxxxx" <xen-api@xxxxxxxxxxxxxxxxxxx>

Hi there,

We've just upgraded to XCP 1.1, and I'm testing Intellicache to see what performance benefits it could possibly bring to our infrastructure. I'm running a Windows 7 VM from an NFS SR, the local storage on the XCP host has caching enabled. So far, so good.

I ran the following command:

xe vdi-param-set uuid=75eaac3e-d2ec-4f45-ac9f-67586f2d703c allow-caching=true on-boot=persist

which enabled caching on the c: of the VM in question. Using dstat on the XCP host, I can see now how when a programme is used data is written to the local SR and pulled from there on next launch. Although the local storage isn't the fastest (I'll be installing an SSD soon), Intellicache clearly makes a difference, as we have a few VMs competing for I/O on that NFS SR. Anyway, I have noticed a couple of oddities; I rebooted the VM now to test boot speed and Windows forced a chkdsk. Many errors are being found. Before the reboot, I noticed my Firefox profile was somehow missing/outdated. Using Previous Versions I managed to restore a working copy.

Once Wiindows is back up and running, I'll bet my outlook .ost file is corrupt. This is exactly the same behaviour I noticed with the unsupported Intellicache in XCP 1.0. I assumed when I tested it back then that the technology was in a testing phase and these sorts of issues would be sorted by the time 1.1 was released. So either there is a problem or I'm doing something wrong (I hope the latter!)

I have a feeling this is related to snapshots; the VM is a member of a VM Protection Policy, which performs a simple disk snapshot nightly at 23:00. But who knows?

So, any clues? I'd appreciate any help working out what's causing this corruption.

Thanks.

_______________________________________________
xen-api mailing list
xen-api@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/mailman/listinfo/xen-api

 

_______________________________________________
xen-api mailing list
xen-api@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/mailman/listinfo/xen-api