In the field of HPC computing, not only CPU computing power is important, IO system data transmission is a bottleneck, high bandwidth, low latency IO is also the key. Intel recently took back the IO500 with a new generation of storage system Wolf, with a read and write bandwidth of 371.67GB/s. Unlike the TOP500 super-counting ranking, IO500 is not very famous, mainly concerned about the IO performance in hPC, last year’s SC19 super-calculation conference, WekaIO with its own WekaIO system won the IO500 first, a total score of 938.95 points, bandwidth 174.74GB/s, performance 5045.33K IOPS.
At the time, Intel’s Wolf system came in second with 933.64 points, but Intel had the advantage of taking only 26 nodes, compared with 345 for WekaIO.
At the ISC20 conference in July, Intel doubled Wolf’s nodes from 26 to 52, and doubled the total score to 1792.98 points, bandwidth 371.67GB/s, and 8649.57K IOPS, or 8.65 million IO performance.
WekaIO’s performance hasn’t been upgraded this year, so it was successfully overtaken by Intel, not just the first, but the second, far ahead of the rest of the system.
Looking at the bandwidth of 371GB/s alone, Intel’s Wolf is not the strongest, with up to 515GB/s bandwidth, but the random performance is far worse, the latter is only 154K IOPS, a difference of nearly 60 times.