24 ms·
Xilinx to Acquire Solarflare
- posnet 8y agoFirst Arista buys Metamako, then nvidia buys Mellanox, now Xilinx buys Solarflare. I wonder who will buy Exablaze?
- chx 8y agoQualcomm to have a better datacenter offering?
- slimscsi 8y ago“low latency networking solutions for customers spanning FinTech to cloud computing.“ Is this an HFT play?
- ianhowson 8y agoHigh speed, low latency and reconfigurable networking are areas that FPGAs are good at. They're already used widely for HFT.
- 1e-9 8y agoIt is certainly a big deal for HFT. Pulling the Solarflare networking technology into a Xilinx FPGA should cut a significant amount of latency for FPGA-based trading systems. Having a Solarflare IP core option should benefit other high-performance networking applications outside of HFT as well. Seems like a smart acquisition for Xilinx.
- willglynn 8y agoIf you want an FPGA tied to a SolarFlare NIC, that's been available for years as a finished product under the name "Application Onload Engine". News article: https://www.theregister.co.uk/2012/02/08/solarflare_application_onload_engine/ https://www.theregister.co.uk/2012/02/08/solarflare_applicat... Whitepapers: * https://solarflare.com/wp-content/uploads/2018/11/SF-114649-CD-LATEST_Solarflare_AOE_SFA7942Q_Product_Brief.pdf https://solarflare.com/wp-content/uploads/2018/11/SF-114649-... * https://solarflare.com/wp-content/uploads/2018/11/SF-114903-CD-LATEST-Solarflare_Application_Nanosecond_TCP_Send_Paper.pdf https://solarflare.com/wp-content/uploads/2018/11/SF-114903-...
- 1e-9 8y agoYes, but the Solarflare ApplicationOnload Engine has several issues (e.g. limited Altera FPGA, inflexible architecture, logic on a separate ASIC that must communicate with the FPGA, and still uses a host CPU for the networking stack), which should be resolved by fully integrating with the Xilinx Ultrascale+ technology and putting all networking and decision logic on one chip. The result should be a much faster, much more flexible, and much more powerful solution.
- scott00 8y agoI doubt it. HFT is a shrinking industry. Solarflare implements their high end NICs using FPGAs, so they have deep expertise in FPGA implementations of network stuff, as well as integrating those implementations with the host OS (ie, writing drivers). My guess is they're interested in offerings related to what's commonly called "software defined networking": running multiple virtual networks over a single physical network, dynamic route optimization, load balancing, etc. This is a key building block of cloud computing, which is a fast growing industry. Implementing it in FPGA firmware instead of actual software makes a lot of sense, so I suspect Xilinx is planning to improve it's offerings in this area.
- ncmncm 8y agoThe absolute number of NICs needed by HFT is quite small, but they have enormous amounts to spend on customization, which can drive product development for others.
- saagarjha 8y agoThey had their earning call today too, didn't they? One of the two seems to have done a number on their stock price.
- xvilka 8y agoProblem with Xilinx is their IDE/toolchain. Hopefully SymbiFlow[1] will take off. [1] https://symbiflow.github.io/ https://symbiflow.github.io/
- lnsru 8y agoWhat problem? Can you imagine some open source toolchain used in corporate setting? Who’s supporting it? And who is liable?
- Benjamin_Dobell 8y agoOpen source and enterprise aren't incompatible e.g. Red Hat.
- lnsru 8y agoIn Xilinx case some chips are also involved. It’s not simple case with software only.
- threeseed 8y agoOpen source is used in every enterprise e.g. Datastax (Cassandra), Confluent (Kafka), Oracle (MySQL), Hortonworks (Hadoop), Databricks (Spark), Everyone (Kubernetes). Vendors support the products but people know it's open source and potentially liable to fail.
- ncmncm 8y agoProprietary software companies are equally likely to fail, and take the source code down with them. Free Software (and, to a lesser extent, Open Source) are less risky than proprietary software.
- Neracked 8y ago> Can you imagine some open source toolchain used in corporate setting? I am working on a multi-billion dollar project and we rely on GCC to build our software. It seems obvious when dealing with "pure" software. It seems unthinkable with "hardware level" software (for FPGA). > Who’s supporting it? Each time we have needed specific expertise, we have paid an external company (Embedded Brains) to provide expertise and updates/fixes. Everything they do ends up being-opensourced. Their expertise is valuable. However, if they disappear tomorrow, we have everything needed to keep the rest of the project going without major disruptions. > And who is liable? I guess every entity involved is liable for its own scope/level. It's not like we will sue the company that provided the toolchain if they are the root cause for a critical anomaly makes it to the end product and nobody noticed that during testing.