Hi, I currently have a database setup with Aerospike but I am lookinng into transferring to Foundationdb along with the document layer. So far I think I have managed to set up a sensible 3 machine cluster (about 128GB per machine) following a lot of advise on this website but it is still significantly slower to write to FDB than it is to Aerospike, by at least an order of magnitude. Basically is it possible to get Foundationdb to run as quickly as Aerospike? Or does anyone have any solid comparisons between the two?
I am not familiar with Aerospike so I don’t know whether your result is expected or not. In order to help, can you please provide some more details about the workload you are running and your configuration?
Also are you sure you’re saturating your cluster? And how large are your transactions? Starting a transaction is relatively expensive in FDB you might be able to get a much higher throughput by writing larger transactions (or by sharing read versions between transactions) - though FDB should do many of these optimizations automatically for you.
And can you share the numbers you’re getting?
Each Machine has 94GB of RAM and 12 cores each. Each machine currently has 12 processes with 3 as tansaction logs, 3 stateless processes, 2 as storage and the remaining 4 without a class asigned.
At the moment I only have test data from my local machine: locally Aerospike is capable of saving 1000 Json files (4KB each) in roughly 900ms but it takes Foundationdb and the document layer upwards of 1000 times longer to complete this when only writing one file at a time. When writing 40 files per transaction FDB is still about 10x slower than aerospike. I will post test data from my 3 machine cluster when I get some.
Basically, does my 3 machine cluster setup look sensible or could I optimise it better?
1 Like