Approached February 2011 with a design for a commodity white box solution to compete with Vblock and vPod.
Designed to run High Density Generic Cloud Workloads as a trial within VMware
Cost per TB primary design consideration but good IOP’s performance important
Architected around Commodity X86 Hardware with our technology partners
Nexenta HA design and scaleout for future with NFS referalls
Classic POD Cloud Design
Approached June 2011 with an Invitation from VMware to use the SuperRack to run some of the Hands On Labs for VMworld 2011.
Although the SuperRack is not designed for the extreme and unique nature of the Hands on Labs. However this represents a massive market opportunity to Nexenta that can’t be missed.
Agreed that due to the late stage in the game that Nexenta will only run 2 out of the 8 verticals in order to field test the SuperRack.
Phased implementation and field testing promised by VMware over the coming months
Ability to run HOL simulation workloads so Nexenta can optimise ahead of the Labs seeing as the other 2 sponsors have run the labs before
All technology for the Labs is Bleeding Edge
New Switches from Arista
VSphere 5
ESXi 5
VCloud Director used instead of Labmanager in previous years
Latest version of VMware View
Nexenta 3.1 RC4
As the “new kid on the block” using a general purpose compute and storage cloud, Nexenta has much, much more work to do in much less time than last year’s Hands on Lab supporting vendors.
Nexenta / VMware form a special team to isolate and fix remaining issues along with Arista over a long weekend. Nexenta Tiger team gets a resource on site in Vegas within 8 hours of launching final troubleshooting and launch.
Nexenta Tiger team work for 70 Hours straight to reverify the entire infrastructure from the ground up with NexentaStor 3.0.5 whilst verifying the entire network infrastructure with CISCO and Arista. Serious Network complexities are resolved and minor Nexenta configurations are optimized and the platform is made stable.
Hands on Labs Real Workload
50 Hours of Lab time during the show
487 Simultaneous Lab seats an hour
2 Hours Lab time on average per attendee
5042 Attendees took the Labs
12,388 Lab Hours sat in total
13,415 Labs Taken by attendees
Nested VM’s workload – Highly latency sensitive
Each lab deploys between 10-25 VM’s per lab seat
Completely Random based choice of labs – No way to determine and precache
Each VM up to 26GB in size
148,103 VM's Created
1 VM created every 1.215 Seconds
By the way – we also destroy them at the end of each lab…..!!!!
How did we stack up?
Ran 4 of the 8 Workloads for the duration of VMworld
During the first session one vendor is offline – We take their workload as well
7 Billion IOPs server from 4 Heads during the labs
3 Billion IOPs server from 1 Controller
38,590 NFS 16K NFS ops on a single controller
Peak controller load 154K 4K NFS ops at sub 1 ms latency
Highest Bandwidth on a Single head (Average I/O 16k)
1,305 MB/sec Total
928 MB/Sec Reads
376 MB/Sec Writes
Less than 2 ms latency
RAM failure in a head triggers HA failover
HA works faultlessly and Controller 3 runs the workload of 2 controllers for 6 hours seemlessly.
Delivered over 50 % of probably one of the most extreme workloads known. In one of the most high profile environments there is in the industry
487 peoples labs offline would have been noticed !
Delivered it on a non optimised generic X86 platform – Enterprise Storage for Everyone
Changed the VMware Infrastructure Engineering Teams view on legacy Storage forever
Made Nexenta more visible to our Competitors and Customers than ever before
Quotes
“ You’re standing toe to toe with the big guys and kicking serious arse”
Dan Anderson – VMware IE team leader VMworld
“ Well done guys it was seamless. We could not tell if it was EMC, NetApp or Nexenta and nor could the customers”
Everett Golden – VMware IE team VMworld
