Looking at the disk LED's it's obvious that it's doing some heavy work - wich I'm guessing is raid5 initialization.

If they're all new drives, your initializations should be abbreviated.  The drives don't have to be zeroed.

How are you 'testing' performance?

Yes, the EMC engineer set it to the recommended values. We also have FAST Cache.

(I can't figure out how to add attachments, sorry for the wall of images)

did you configure/enable system cache ?

Include a description of your pool's physical (drives: number, type and speed)  and logical (RAID 5 LUNs: number, size, etc.) construction.

Have you looked over: Proven Solution Guide: EMC Infrastructure for Virtual Desktops - Enabled by EMC Celerra Unified Storage (FC), VMware vSphere 4.1, VMware View 4.5, and VMware View Composer 2.5 ?  Available on Powerlink.

The pool consists of 15x600GB 10k SAS drives and 5x2TB 7200rpm NL-SAS drives. Auto-tiering is enabled. I've verified that the LUN in question is allocated 100% on the fastest tier.

There is only one 1TB LUN created, the rest is unused.

I haven't read that guide, as we're not using the NAS side (yet), or vmware view. I will however take a look.

I haven't done any real testing so to speak, we just migrated a few test VM's over and observed how the systems behave. Generally, booting takes almost 2 minutes longer than before, general application launch times are slower - applications often hangs (waiting for I/O is my guess) etc.

The only "benchmark" I've done is ATTO diskbench, wich shows horrible performance with small blocks(the numbers on the left and right side of the charts are in kilobytes/sec, while the charts shows MB/s).

I've also added the same test from a VM running on the same ESXi host against a old AX4-5i SAN with raid5 15k.

VNX5300:

AX4-5i:

As you clearly can see, the AX4-5i generally beats the living.. out of the VNX.

OK.  This is explainable.

If you get a chance, go over EMC Unified Best Practices for Performance and Availability: Common Platform and Block O.E. 31.5.  The guidance for pools has changed a bit.

I assume you used the default 'Auto-tier' setting for the pool.  The recommended setting is 'Highest'.

A single LUN is not ideal for a pool.  You should have a minimum of two, but more is better.  Efficency begins to scale steeply upward with four and more.

I think your cache is 'unwarmed'.  That is, try the boot exercise at least four times, or with > 3 boxes twice. The 'chunks' need to be promoted into the cache.

HTH

You may find this thread helpful: https://community.emc.com/thread/124312

jps00 wrote: OK.  This is explainable. If you get a chance, go over EMC Unified Best Practices for Performance and Availability: Common Platform and Block O.E. 31.5.  The guidance for pools has changed a bit.

John,

where does one find this updated paper ?

Thanks

I don't understand.. why should it matter if I have 1 LUN or 4 LUN's across the pool? Doesn't it stripe the data along all the raid groups/disks anways? Remember that this is just initial testing/lab'ing, we don't have enough horsepower to actively benchmark 4 LUN's at the same time without degrading the performance of our production systems.

I know that the recommended setting for high-activity data is to use 'Highest' tiering - but like I said, the data is already on the highest tier (verified via unisphere, 100% data on highest).

Regarding FAST Cache: Yeah, thats probably right. But even with out FAST Cache the boot speed should not be 3x the time it takes on a old AX4-5i without FAST Cache..