ZFS Question on intent log

P

pillai_hfx

I have two 45 drive ZFS file servers on two different clusters. The SLOG device used in both cases is the Intel DC S3700. I did two IOR benchmarks across many NFS clients in parallel using MPI with two different logbias settings for the file systems.

One was latency and other was throughput.

With latency, I could see that the SLOG device is getting used, while in the second case it is not. The write speed is the same! The sync is set to standard on both file systems. Both setups have four RAIDz3 vdevs with 11 drives each.

So my question is, does having a separate intent log help on larger pools on file server duty? Thanks.
 
Here is an interesting link I found that asks the same question. https://github.com/zfsonlinux/zfs/issues/1012

When the pool throughput is much higher than a single solid state drive's throughput and provided that ZFS is used in a non latency sensitive environment like a file server for example, wouldn't adding anything other than a ddrdrive (like a normal SSD) in fact hurt the write throughput instead of helping?

Same with L2ARC. On one of our clusters, the utilization is just over 50% and on the other one it is just 3%!. So unless the file server gets lots random read workload, even L2ARC is not very beneficial it seems. All it might do is to create more write overhead and waste some ARC space.

I am planning a bigger deployment with 180 drives next year subject to funding approval. That is why I have these questions about benefits provided by SSD to ZFS. If they are not very beneficial, I can concentrate on simply getting more processor cores and memory for the server within the budget instead of wasting money on these SSD drives. Thanks.

[ capitalized acronyms like SSD and ZFS don't need additional [file] tags -- Mod. ]
 
You must log in or register to reply here.
Back
Top