Infinidat benchmarking beatdown: Glugging slimfastq? Not us!
Bottles benchmark battle backlash
Analysis Infinidat recently ran a benchmark in which it claimed to have beaten EMC and Pure systems.
We spoke to its rivals as well as research sources in the industry, and more than one source contested the results, claiming they had found out a few things about the test. They suggested that the Infinidat system was unrealistically configured and spoke of a special, memory-enhancing mode and provided us with some data.
So we contacted Infinidat with questions about it.
Firstly, we said we understood that the test configuration featured:
- F6230 with 3x active:active:active controllers (servers) with InfiniBand interconnectivity
- 1.1TB of DRAM/controller and 3.3TB in total
- 200TB of SSD second-level cache
- Compression was enabled
- 42U of rackspace
As Infinidat currently ships 1.9TB SSDs, it appears that the maximum shippable SSD cache capacity is 100TB. Does that mean that, for the test, Infinidat used 3.8TB SSDs, which it is not yet shipping?
CTO Brian Carmody said: "The test configuration is fully documented in our blog post. Unlike hardware-centric... solutions, we do not specify nor require specific drive sizes, only total capacity, and yes, our public data sheet is, by design, conservative."
Would a typical configuration array (F6230 with 1PB of storage, 24TB of SSD cache and 1TB of DRAM) deliver the same results as the tested system?
Carmody said: "We confirm that InfiniBox with 24TB of SSD will beat all-flash Unity in real life and costs much less."
The Unity system EMC used in its tests featured:
- 2 active:active controllers
- About 660MB DRAM
- SSD used as storage tier
- 2U of rackspace
We might suggest that it is hardly surprising the F6320 beat the Unity system – it's like comparing a Ferrari to a pick-up truck. What could Infinidat say on that point?
Carmody said: "Of course we are not surprised. Our customers have deployed 1.5 Exabytes of InfiniBox capacity running core banking workloads migrated off VMAX, Splunk migrated off VNX/Unity, cloud file systems migrated off Isilon, and disaster recovery workloads migrated off Data Domain."
It went on to claim that "EMC needs a vast, complex portfolio, including five new product announcements this week, to compete with one Infinidat product", adding that, in its view, "the future is about massive pools of universal storage per data center".
Meanwhile, on the compression front, our sources claimed the InfiniBox can be set to have several compression levels, and that there is a "competition mode" in the system – and pointed us to GitHub to indicate that was so.
Lines 157-160 read:
1: Uses less than 4M memory (!), yields the worse compression (still much better than gzip)\n\
2: Uses about 30M memory, reasonable compression \n\
3: Uses about 80M memory, best compression <default level> \n\
4: Compress a little more, but very costly (competition mode?) \n\
So... was competition mode enabled in the test, when it typically would not be in real-world InfiniBox usage?
Carmody denied it. He replied: "There are no levels of compression on InfiniBox – it is enabled by default for all entities. There is no reason to disable compression other than to prove it has no performance impact. The compression function... found on our GitHub page is from one of our many (30) open-source projects on GitHub – which is a utility that compresses genomics FASTQ files that we developed three years ago.
"The slimfastq utility is not part of InfiniBox and has no relevance here. We suggest [folks] spend less time on competitive intelligence and more time driving cost out of their bloated portfolio."