The default erasure plugin in Ceph is the jerasure plugin, which is a highly-optimized open source erasure-coding library. The library has a number of different techniques that can be used to calculate the erasure codes. The default is Reed-Solomon, and this provides good performance on modern processors, which can accelerate the instructions that the technique uses. 

There are also a number of other jerasure techniques that can be used, which all have a fixed number of M shards. If you are intending on only having two M shards, they can be a good candidate, as their fixed size means that optimizations are possible, lending to increased performance. Each optimized technique, aside from only supporting two erasure shards, also tends to have certain requirements around the total number of shards. Here is a brief description of each optimized technique:

• reed_sol_van: The default technique, complete flexibility on number of k+m shards, also the slowest.

• reed_sol_r6_op: Optimized version of default technique for use cases where m=2. Although it is much faster than the unoptimized version, it's not as fast as other versions. However, the number of k shards is flexible.
• cauchy_orig: Better than the default, but it's better to use cauchy_good. 
• cauchy_good: Middle-of-the-road performance while maintaining full flexibility of the shard configuration.
• liberation: Total number of shards must be equal to a prime number and m=2, so 3+2, 5+2, or 9+2 are all good candidates, excellent performance.
• liber8tion: Total number of shards must be equal to 8 and m=2, only 6+2 is possible, but excellent performance.
• blaum_roth: Total number of shards must be one less than a prime number and m=2, so the ideal is 4+2, excellent performance.

As always, benchmarks should be conducted before storing any production data on an erasure-coded pool to identify which technique best suits your workload.

In general, the jerasure profile should be preferred in most cases, unless another profile has a major advantage, as it offers well-balanced performance and is well-tested.