We measure capture probes by primarily looking at the following metrics:

• Sustained capture rate (measured in bits per second of bandwidth)

• Sustained capture packet rate (measured in packets per second)

We measure these in isolation per PMUX and then we look at how many PMUXes each hardware variant can support concurrently. We do additional testing to confirm that each capture PMUX scales independently and does not interact with other capture PMUXes running on the same appliance.

The capture probe performance benchmarks assume that only VMX-Capture components are running on the hardware.

The capture rate performance benchmarks are affected by both Capture Probe/PMUX software performance, and also appliance hardware configuration. By contrast, the ‘capture storage (compressed)’ metric is determined primarily by the appliance hardware and by the compressability of the data.

All-in-one appliances versus dedicated Capture appliances

Note that to achieve the higher levels of sustained capture rate, the appliance performing the capture needs to be a dedicated capture server with VMX-Analysis running on separate hardware. The following table maps between the all-in-one appliance models listed in Performance Benchmarks and Benchmark appliances to the equivalent dedicated capture model:

The use of dedicated capture appliances allow capture performance to be split between two different RAID controllers, raising the level of capture performance that is possible from that appliance.

Napatech Capture buffer

These benchmarks were produced using a standard Napatech capture buffer (4GB). A bigger capture buffer will absorb more bursty traffic (which will allow for a greater Sustained capture rate), but it will reduce the amount of RAM available on the appliance for packet multiplexers and databases (see below).

Compression cards to improve capture rate

We have already seen how running multiple PMUXes in parallel can achieve high capture rates, and how increasing the Napatech capture buffer value will provide even better performance when dealing with bursts of traffic.

Beeks Analytics can also use one or more FPGA compression cards to improve capture rates. These cards compress data on-the-fly before it is written to disk, without any impact on the server CPU. Each compression card can support a compression rate of around 60Gbps.

Beeks is a flexible solution and supports custom specifications, including increasing the number of compression or removing compression cards altogether(if sufficient PCI slots are available in the appliance). Three compression cards installed concurrently in an appliance have been demonstrated to produce capture rates of 180 Gbps and above.

For even faster write times, or where it is not important to compress the capture files, it is also possible to write uncompressed to the disks as well. We have typically observed sustained capture write performance for each RAID controller to be 80 Gbps. The uncompressed write times of the appliances are therefore as follows:

These numbers can be improved upon up to around 100 Gbps per RAID controller or greater, but that is more heavily dependant on disk performance and the specific workload.

See Benchmark appliances earlier for details of how many compression cards are assumed to be installed in each appliance, and see ‘Capture storage (compressed)’ below for a further discussion of compression ratios for different traffic types.

Compression ratio

The numbers provided assume roughly 3:1 compression (we have observed that this is typical for market data workloads). We have observed compression rates as high as 5:1, which will result in proportionately more usable storage.

3:1 compression ratio has been observed to be more typical for market data, 5:1 compression ratio has been observed to be more typical for trading traffic.

Capture storage (compressed)

These figures are the usable storage, based on the typical compression ratios that we observe for trading and market data. See the above discussion of compression cards and compression ratio.

Higher volume capture storage can also be provided by larger disks, or by moving to a dedicated capture appliance, which allows for more flexibility to scale the appliance disks out (see above for equivalent dedicated capture models).