SC07: Y’all remember we’ve talked about networking starting Woven Systems a few times recently (e.g., here). This company is really interesting; their main claim to fame right now is an adaptive network switch that let’s you build clusters with up to 4,000 servers while staying in layer 2.
The company has a couple of new things to talk about this week. You may recall that Woven and Sandia announced some interesting test results that show their technology outperforming InfiniBand in a saturated network test in a 128-node cluster. Woven has updated these results, and you can find a white paper at their web site (here).
Today Woven has announced customer number 2: Sandia (the name of customer number 1 hasn’t been released). I guess this isn’t a surprise, but it is a testament to the satisfaction of a customer that’s worked with the gear for a while in their own environment. From the release
…[Sandia] is deploying the EFX 1000 Ethernet Fabric Switch in a 128-node compute cluster. Sandia chose the EFX 1000 based on its proven ability to sustain 10 Gbps Ethernet throughput in a large remote direct memory access (RDMA) cluster during exhaustive testing at its Livermore, California laboratory.
Of less direct interest to HPC is their new TRX100; it’s a top of rack switch with 48 1GbE ports and 4 10GbE upload links that sports Woven’s adaptive routing technologies. This little beastie is aimed at the scale out datacenter.
So now that they’re announcing customers and new products, how does Woven intend to move forward? I asked during a conversation with Derek Granath, VP of Marketing for Woven. Derek indicated that Woven is working on direct sales right now, but will be looking for vendor partners to incorporate Woven’s switch into their clusters. This will be really important for certain high-end customers (like the DoD), and the mid-market, since both groups like to buy machines that are already integrated and ready to go.