isilon and vmware

[Aglidic]

Hello guys,
i know that isilon and random pattern it's not the best combo but i want to know if some of you run vmware or oracle databases on a isilon cluster.

And if yes on what kind of cluster (node types and number of nodes) and of course if you are happy with the performance?

Thanks

[Andrew Stack]

well to take what will probably be a long thread and sum it up for you in a few sentences:

VMWare - As good as any other NAS solution in terms of performance.  It's all about your design in that sense.  The big differential between Isilon and other NAS vendors is its sotware offerings for VMware.  The plugin for VCenter is feature anemic, no dedupe (coming), and you're basically restricted to NFS (which I personally like but some folks insist on LUN's).  So, it's supported and workable you just don't get the added software benefits when compared to it's peers.

Oracle - In the words of a officer Barbradey.  "Move along people, nothing to see here".

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-gr...@googlegroups.com.
For more options, visit https://groups.google.com/groups/opt_out.
 
 

--

Andrew Stack

Sr. Storage Administrator

Genentech

Cell - 650.867.5524

[Aglidic]

Thanks
yeah it's nearly what i was thinking so no databases on isilon and for vmware i will say s200 with ssd required (only for metadata).
So we will wait oneFS 8 and new optimisation before putting random pattern on isilon.

[Cory Snavely]

Anybody out there with a preference for basing VMs on LUNs?

On 6/18/2013 4:41 PM, Aglidic wrote:
> Thanks
> yeah it's nearly what i was thinking so no databases on isilon and for
> vmware i will say s200 with ssd required (only for metadata).
> So we will wait oneFS 8 and new optimisation before putting random
> pattern on isilon.
>

> Le mardi 18 juin 2013 22:37:45 UTC+2, Andrew Stack a �crit :

> <javascript:>.

> For more options, visit https://groups.google.com/groups/opt_out

> <https://groups.google.com/groups/opt_out>.

[Aglidic]

I prefer NFS but in Europe (france/switzerland) most of the time people are like afraid of NFS for vmware and always used FC just can't undderstand why

Le mardi 18 juin 2013 22:58:32 UTC+2, Cory Snavely a écrit :

Anybody out there with a preference for basing VMs on LUNs?

On 6/18/2013 4:41 PM, Aglidic wrote:
> Thanks
> yeah it's nearly what i was thinking so no databases on isilon and for
> vmware i will say s200 with ssd required (only for metadata).
> So we will wait oneFS 8 and new optimisation before putting random
> pattern on isilon.
>

> Le mardi 18 juin 2013 22:37:45 UTC+2, Andrew Stack a �crit :

[Cory Snavely]

Well, presumably FC would be preferable because of the lowered latency
of SCSI over FC as opposed to NFS over TCP over Ethernet. I'm curious
about people who prefer iSCSI on Isilon clusters over NFS on Isilon
clusters for VMs...or maybe I misread the earlier post.

On 06/18/2013 05:09 PM, Aglidic wrote:
> I prefer NFS but in Europe (france/switzerland) most of the time people
> are like afraid of NFS for vmware and always used FC just can't
> undderstand why
>

> Le mardi 18 juin 2013 22:58:32 UTC+2, Cory Snavely a �crit :

>
> Anybody out there with a preference for basing VMs on LUNs?
>
> On 6/18/2013 4:41 PM, Aglidic wrote:
> > Thanks
> > yeah it's nearly what i was thinking so no databases on isilon
> and for
> > vmware i will say s200 with ssd required (only for metadata).
> > So we will wait oneFS 8 and new optimisation before putting random
> > pattern on isilon.
> >

> > Le mardi 18 juin 2013 22:37:45 UTC+2, Andrew Stack a �crit :

[Jason Davis]

Not sure if there is an advantage really, unless you like working with block devices :p

For general low IO usage in VMWare, the Isilon works good. Especially in 7.x, NFS write latency is pretty decent, consistently 4ms or so. As for something really IO intensive, like a Large relational database my suggestion would be to look at local storage or a more traditional SAN.

Below is a Ganglia VM we migrated over to our IQ108NLs cluster running OneFS 7.0.1.? (at the time)

On Tue, Jun 18, 2013 at 4:20 PM, Cory Snavely <csna...@umich.edu> wrote:

Well, presumably FC would be preferable because of the lowered latency of SCSI over FC as opposed to NFS over TCP over Ethernet. I'm curious about people who prefer iSCSI on Isilon clusters over NFS on Isilon clusters for VMs...or maybe I misread the earlier post.

On 06/18/2013 05:09 PM, Aglidic wrote:

I prefer NFS but in Europe (france/switzerland) most of the time people
are like afraid of NFS for vmware and always used FC just can't
undderstand why

Le mardi 18 juin 2013 22:58:32 UTC+2, Cory Snavely a écrit :

    Anybody out there with a preference for basing VMs on LUNs?

    On 6/18/2013 4:41 PM, Aglidic wrote:
     > Thanks
     > yeah it's nearly what i was thinking so no databases on isilon
    and for
     > vmware i will say s200 with ssd required (only for metadata).
     > So we will wait oneFS 8 and new optimisation before putting random
     > pattern on isilon.
     >

     > Le mardi 18 juin 2013 22:37:45 UTC+2, Andrew Stack a ďż˝crit :

an email to isilon-user-group+unsubscribe@googlegroups.com.

For more options, visit https://groups.google.com/groups/opt_out.

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.

To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-group+unsubscribe@googlegroups.com.

[Charles Llewellyn]

We are using Isilon X200 with SSD for metadata to provide tier 2 storage to back our vCloud director platform. Initially we didn't have SSD in the equation and had some significant problems with latency. Introducing SSD reduced latency by half for random write operations and doubled the number of IOPs we could push from a single VM. So whatever you consider you really need SSD. Also from experiences I would recommend only considering the S-series as the faster disks will make a significant improvement to latency.

A word of warning though we have recently been undertaking some benhcmarking to evaluate version 7 and have noticed a huge reduction in performance. We have yet to get to the bottom of the issue but it looks like it could be a combination of linked clones on vSphere 5.0 which doesn't allow vendor to set the delta block size causing mis-aligned writes negating the benefit of endurant cache and the new disk layout reducing the number of spindles available to stripe VMDKs across. 

That said if you are starting from scratch using 7.x with vSphere 5.1 the integration with VMware via VAAI should negate these issues.

If you do evaluate it I would be interested in your finding :)

Cheers

Charlie

[Richard Kunert]

We run a 6 node VMware cluster against a 7 node 72000x cluster (2 TB SATA drives - no SSD, but 250+ spindles) and it works just fine. We also have an 18 node XenServer cluster that runs Sun Grid Engine and assorted bioinformatics software using the Isilon as the backend. No particular performance issues.

We run a pretty large Oracle database against the Isilon and that works very well too. Oracle has something called direct NFS where Oracle itself makes the NFS connections. Typically we see our Oracle server (which is a VMware VM) making about 100 connections to the cluster.

It really all depends on your workload. I'm a big proponent of making as many measurements as possible before speccing out your hardware (munin is great if you're running Linux). The Isilon isn't the greatest in terms of latency but our cluster can push about 1000 IOPs on a GbE connection - and we have 24 GbE interfaces. We also have a couple of 10 GbE interfaces but so far we're only using those for NDMP backups of the cluster.

--Richard

--
You received this message because you are subscribed to the Google Groups "Isilon Technical User Group" group.

To unsubscribe from this group and stop receiving emails from it, send an email to isilon-user-gr...@googlegroups.com.

[Aglidic]

Ok guys thanks for your answer, it seems they are on a good way to prive a NAS who can support Random I/O pattern.
I hope oneFS v8 will do that as well as zfs or netapp.