[Preamble: I was a delegate of Storage Field Day 15 from Mar 7-9, 2018. My expenses, travel and accommodation were paid for by GestaltIT, the organizer and I was not obligated to blog or promote the technologies presented at this event. The content of this blog is of my own opinions and views]
My conversation with Huawei actually started the night before our planned session at their Santa Clara facility the next day. We had a evening get-together at Bourbon Levi’s Stadium. I was with my buddy, Ammar Zolkipli, who coincidentally was in the Silicon Valley for work. Ammar is from Hitachi Vantara Japan, and has been a good friend of mine for over 17 years now.
Shortly, the Huawei team arrived to join the camaraderie. And we introduced ourselves to Chun Liu, not knowing that he is the Chief Architect at Huawei. A big part of that evening was our conversation with him. Ammar and I have immersed in the Oil & Gas EP (Exploration & Production) data management and petrotechnical applications when he was in Schlumberger and after that a reseller of NetApp. I was a Consulting Engineer with NetApp back then. So, the 2 of us started blabbering (yeah, that would be us when we get together to talk technology).
I observed that Chun was very interested to find learn about real world application use cases that would push storage performance to its limits. And I guessed that the best type of I/O characteristics would be small block, random I/O and billions of them, with near-real time latency. After that evening I did some research and could only think of a few, such as deep analytics or some applications with needs for Monte Carlo simulations. Oh, well, maybe I would share that with Chun the following day.
The moment the session started, it was already about the speed prowess of Huawei Storage. It was like the greyhounds unleashed going after the rabbit. In the lineup, the Dorado series stood out.
And Huawei was very proud to show their speed pedigree in the SPC-1 v3 benchmark (June 2017). Here were the results of that benchmark.
The highlight of the session with Huawei was Chun Liu going deep dive with their storage architecture. And the core of its high performance architecture was the design thinking that was baked into its DNA. Several points stood out:
- A non-blocking, shard to core affinity, lightweight threading architecture
- Shared-nothing scale out architecture
- An optimized data path architecture which involved optimized LZ4 implementation and SIMD (Single Instruction Multiple Data) vectorization for parallelization
- Simplified and optimized metadata management methods using pre-compressed B+ tree, block addressing and fingerprint indexing and mapping
- Optimized ROW (redirect-on-write) and aggregation techniques, even for triple parity RAID
- Understanding SSD layout for optimized Garbage Collection (GC)
- Multi-Stream Writes (MSW) SSDs
- Optimized in-line Global Deduplication
- Many more …
One table from Chun’s presentation impressed me a lot. Look at the red rectangle … hardly a whisper even with dual or triple parity RAID. Impressive!
There is plenty to look forward to with Huawei Storage, especially the Dorado series. It was engineered to be a speed demon, a performance beast!
And we know that Huawei is pushing the envelope for higher and higher speed. There are rumours in the market about their next generation storage, one that can scale to 16-nodes, 256 blades, 12,288 CPU cores, 32 million IOPS and 200 microsecond latency. Something code named Barreleye.
The inaugural field day session with Huawei Storage left us with a strong desire for speed. More and more speed! And we do hope we can see a Barreleye in the future. What say you, Huawei?