So if you like to see this project to continue and get official support inside a EMC, please go on and fill a EMC Symmetrix VSA Questionnaire at http://storageboy.com/2011/08/29/vmax-vsa-its-alive-2/

NexentaStor is available in two flavors, Community Edition (later CE) is free for non-production use and supports up to 18 TBs of data. For production use there is Enterprise Edition which is licensed by required capacity. VAAI is supported on both editions which makes NexentaStor CE very good candidate for vSphere home lab storage appliance.

I have been running NexentaStor on my home vLab for some time now so implementing VAAI was just a simple NexentaStor appliance upgrade task. Once appliance was upgraded to latest 3.1.1 release VAAI hardware acceleration was immediately available to my ESXi 5.0 host:

Below is a copy & paste from NexentaStor 3.1.1 Release Notes explaining available VAAI primitives:

1. SCSI Write Same:  When creating a new virtual disk VMware must write‟s zeros to every block location on the disk. This is done to ensure no residual data exists on the disk which could be read by the new VM. Without Write Same support the server‟s CPU must write each individual block which consumes a lot of CPU cycles. With the Write
   Same command VMware can direct the storage array to perform this function, offloading it from the CPU and thereby saving CPU cycles for other operations. This is supported in ESX/ESXi 5.0 and later.
2. SCSI ATS: Without ATS support, when VMware clones a VM it must lock the entire LUN to prevent changes to the VM while it is being cloned. However, locking the entire LUN affects all other VM‟s that are using the same LUN.  With ATS support VMware is able to instruct the array to lock only the specific region on the LUN being cloned. This
   allows other operations affecting other parts of the LUN to continue unaffected. This is supported in ESX/ESXi 5.0 and later.
3. SCSI Block Copy: Without Block Copy support, cloning a VM requires the server CPU to read and write each block of the VM consuming a lot of server CPU cycles. With Block Copy support VMware is able to instruct the array to perform a block copy of a region on the LUN corresponding to a VM. This offloads the task from the server‟s CPU
   thereby saving CPU cycles for other operations. This is supported in ESX/ESXi 5.0 and later.
4. SCSI Unmap: Provides the ability to return freed blocks in a zvol back to a pool.  Previously the pool would only grow. This enables ESX to destroy a VM and return the freed storage back to the zvol. ESX 5.0 and later support this functionality

VAAI performance impact

To see what impact VAAI has on my vSphere lab I ran some test by deploying a VM from template. Test template was a Windows 2008 R2 VM having 40 GB thin disk with 10,4 GB of data and 4 GB VM swap file, I ran VM deploy three times for both tests.

Shortest time to deploy VM without VAAI was 488 seconds, with VAAI enabled it took only 248 seconds at worst to deploy a VM. NexentaStor 3.1.1 with VAAI is almost twice as fast to deploy VMs than without!

Test setup

For full disclosure here is test setup used. Please note that some parameters shown here are not safe for production use!

ESXi host had two dedicated iSCSI connections to NexentaStor appliance over quad port Intel PRO/1000 adapter. vSphere PSP was set to Round Robin and IOOperation Limit was set to 10 on all devices. MTU on iSCSI connections was set to 9000 on both ends.  NexentaStor was configured with two iSCSI target portal groups which both were configured for single iSCSI target.

VM template was stored on a VMFS datastore cluster with four VMFS 5 datastores, new VM was deployed to same datastore cluster using automatic VMDK placement by Storage DRS.

LUN configuration on NexentaStor for VMFS 5 datastores was compression=on, checksum=off, deduplication=off, log bias=throughput and sync=disabled.

NexentaStor appliance preference changes from default was sys_zfs_nocacheflush=yes, sys_zfs_vdev_max_pending=1, Net_tcp_naglim_def=1, net_tcp_recv_hiwat=524288 and net_tcp_xmit_hiwat=524288.

ESXi host software and hardware

• VMware ESXi 5.0 build 469512
• Intel Core i5 2500K CPU
• 16 GB DDR3-1600 RAM
• Intel PRO/1000 Quad Port Ethernet adapter

NexentaStor appliance software and hardware

• NexentaStor 3.1.1
• Intel Core 2 Duo E8400 CPU
• 8 GB DDR2-1066 RAM
• Intel PRO/1000 Quad Port Ethernet adapter
• 6 x Seagate 7200.11 500 GB SATA drives in RAIDZ-0

VSP features automatic storage tiering on sub-LUN level much similar to EMC FAST on CLARiiON, HDS calls this as Hitachi Dynamic Tiering or HDT. By default all data is stored in highest available tier, HDT will move data down in tiers as it is accessed less frequently. HDT is capable of moving data in 42 MB continuous pages, which is same page size used with HDS dynamic (thin) provisioning.

Like USP today, VSP does not support VMware VAAI at launch time but I expect VAAI support to arrive for USP and VSP in Q1 of 2011.

HDS VSP is also available from HP as P9000.

Read more of VSP at HDS.com
http://www.hds.com/products/storage-systems/hitachi-virtual-storage-platform.html

HDS press release
http://www.hds.com/corporate/press-analyst-center/press-releases/2010/gl100927b.html

HP press release of P9500
http://www.hp.com/hpinfo/newsroom/press/2010/100927b.html

I double checked VMware HCL and adapter I had, Qlogic 8100 (QLE8152 to be exact), is listed as supported on ESX 4.1 with ethernet driver qlge 1.0.0.39 and FCoE driver qla2xxx 831.k1.23vmw, I had correct drivers so I filed support case to VMware.

While waiting for a reponse from a VMware support I started digging internals of driver zip. Inside of any VMware patch file is a metadata.zip which hosts, like name states, metadata for that particular patch in XML format. I focused my investigation on vmware.xml and in that I noticed something odd, bundle softwarePlatform version was tagged as 4.* but actual driver was tagged to be compatible with 4.0.0. Below is a snippet of XML blocks in question.

<platforms>
 <softwarePlatform locale="" productLineID="esx" version="4.*" />
 <softwarePlatform locale="" productLineID="embeddedEsx" version="4.*" />
</platforms>

<systemReqs>
 <swPlatform locale="" productLineID="esx" version="4.0.0" />
 <swPlatform locale="" productLineID="embeddedEsx" version="4.0.0" />
 <maintenanceMode>true</maintenanceMode>
</systemReqs>

Feeling adventurous I replaced version="4.0.0" with version="4.*" in <systemReqs> block, zipped all files back together and transferred driver zip files to ESX service console.

With a bit excitement I ran esxupdate for FCoE driver, just to get message that bundle is obsolete as newer package is already installed. Quick "rpm -qa | grep qla" revealed that indeed there is a 831.k1.28.1-1vmw included in ESX 4.1. So ESX 4.1 actually supports QLogic FCoE out of the box, while working on service console only and relying on VMware HCL that additional drivers are required I had not noticed that FC adapters were already visible.

FCoE solved I ran esxupdate for remaining QLogic 10 GbE driver which this time passed successfully, after a reboot I also had two QLogic 10 GbE adapters visible!

I reported my findings to VMware support  and they came back with confirmation that driver should have been working with ESX 4.1 and that they have reported this error back to ESX engineers for driver fix. So, until there is fixed QLogic 10 GbE CNA driver available from VMware you can fix the driver with instructions above, but please do that with your own responsibility!

Features shown in video are

1. New CLARiiON and Celerra management GUI called Unisphere, based on Adobe Flex technology
2. Sub-LUN tiering – “VMware DRS” equivalency for storage medium, will migrate data based on usage between Flash, FC and SATA
3. Flash cache – AKA FAST cache, up to 2 terabytes(!) of 2nd level read/write cache
4. Block level compression – told to save 30% to 50% on average on your storage consumption

It is not told directly in this video but it looks like thin pools will become an standard feature on CX4, at the moment is is sold as an add-on option.

These and new vStorage API integration features make me feel that this is a good time to be EMC customer.