ZFS Seagate IronWolf 110 SSD for boot and SLOG

[gpw928]

I need a 240 GB 2.5" SSD suitable for a boot disk. It will be mirror'd with my old 240 GB Intel SSD DC S3520, and have a small partition for a SLOG.

Does anyone have experience of the Seagate IronWolf 110 SSD? It looks to me to be on product run-out, at just about half the price of a 2.5" Intel DS-S4510 the same size.

This brochure says of the IronWolf 110 "Power is stored in integrated capacitors that immediately supply energy when the power drops below a specific threshold. This emergency supply of energy allows in-flight writes to complete and buffer data to copy to NAND". That seems fairly definitive that it's OK for a SLOG.

The technical specifications suggest it has performance reasonably compatible with it's mirror paired Intel SSD (which is all it has to do).

It looks fit for purpose to me. Any experience with this SSD, or other comments?

 

[Deleted member 67440]

In fact no, seems way faster (bandwith) but with very high latency
Two very old design, in fact, and with TLC (not so good for a not-new media)

Seagate IronWolf 110 SSD NAS Review

The Seagate IronWolf 110 SSD, though announced at CES back in January, only recently hit…

www.storagereview.com

I do not suggest dissimilar mirror drives

 

[ralphbsz]

Both are older models, with decent performance. For a home server or light commercial workload, they'll probably do excellently.

The Seagate with its older TLC has decent endurance: about 2000 write cycles. Barring extreme workloads, that's more than sufficient for a lightly used server.

As fcorbelli hinted at: If you mirror two dissimilar drives, write performance will be the minimum of the two. And that's not "on average", but for each IO, which means that with a mix of IOs, the write performance can be less than half of either drive. Note that this applies to write performance; for read performance, the typical answer is somewhere between the average of the two and the sum of the two, depending on how read scheduling is done.

On the other hand, typical SSDs are so fast compared to small (home?) server workloads that a drop in performance probably doesn't matter. In particular if the alternative is unrealistic. Sure, you could buy two spinning disks ... which are even slower. And if you can't afford a matched pair of shiny, new high-performance SSDs, it doesn't matter that they would be faster.

The boot SSD on my home server is also an Intel, by now about 8 years old, and has been powered on nearly continuously for 6.5 years. I just looked: Total writes are already 731 times user capacity, and it's not showing any sign of failure. That would be excellent for a modern drive.

 

[gpw928]

My initial thought was to get a pair of M.2 NVMe SSDs. However small SSDs with full Power Loss Protection seem to be hard to find at the moment (not helped when PLP doesn't always mean what it suggests). Perhaps it's a COVID-related supply chain problem. Maybe it's tied up with the sale of Intel's NAND business? Or both?

I just found this brochure on the IronWolf 110. See page 9. It seems that their superior performance figures are based on achieving lossless compression of all data by at least 20%. Not what anyone with ZFS LZ4 enabled is going to achieve... I thought it was a bit too good to be true.

Poor random writing with high latency is a disappointment. My old DC S3520 has a quarter of the latency and twice the random write throughput (for compressed data).

I'm was thinking that a D3-S4510 is my best option at this time. It's 30% more expensive than the IronWolf 110, but it is more likely fit for purpose. But, 240 GB Intel D3 SSDs of all kinds appear to be unprocurable.

The Seagate Nytro 240 GB (medium endurance) drive looks OK. I'm assuming that random write latency is directly correlated with Sustained Random Writes (IOPS). They, at least seem to be procurable.

Edit: this spec sheet for the Seagete Nytro 240GB shows that it also relies on data compression for superior performance, and, with uncompressible data, it under-performs my old DC S3520. Sigh...