DS923+ NVMe cache options

[itsjasper]

I'm running Samsung 970 EVO Plus 1TB & 2TB drives in all my NASes, Using the 2TBs as R/W cache on the DS1621+/DS1522+ and the 1TBs are a storage volume in the DS720+ using Daves' scripts... All are on APC UPS units. No failures or issues here.

 

[Searcher214]

I think I'm settled on firing up Cache Advisor, now that I finally have most of our files copied onto this new NAS. If it recommends read only cache, then I'm fine with these FireCuda 530's, as they have 0.7 DWPD, and others have tested and reported compatibility. My only concern was really the possibility of lost data if used as read/write cache, if perhaps a I or a cleaning person knocks the cord loose on the back of the NAS itself, or if the NAS power supply or CPU fail.

@Telos said that increasing RAM might negate any benefit of setting cache to read/write, eliminating any concern over power loss protection. But I haven't seen any support of that in the Synology literature.

Either way, I'll be running the cache RAID-1 if it's set to read/write. Does this decrease the probability of volume corruption during a power failure?

And yes... the NAS is on a nice big UPS with USB comm's to the NAS (well, still to old NAS at the moment), so "power fail" does not mean site power, but local NAS cord/power supply issue.

 

[Telos]

@Telos said that increasing RAM might negate any benefit of setting cache to read/write,

That's not what I said, nor inferred.

 

[Searcher214]

That's not what I said, nor inferred.

I must have misunderstood you, then. I was referring to this:

ICBW, but in terms of saving Excel files more rapidly, you would be better served by increased RAM. NVMe cache better serves heavy database activity, where data is constantly hammered thousands of times in short duration. I suggest 16-32 GB memory.

 

[Telos]

I must have misunderstood you, then. I was referring to this:

That's my post, but your inference extends beyond what I stated.

 

[Searcher214]

That's my post, but your inference extends beyond what I stated.

Sorry about that. Can you elaborate? How does excess RAM impact best choice for cache arrangement?

 

[Telos]

Let's say I'm a typical SOHO user with a 100 € budget... I can afford either cache or memory upgrade. I would realize greater utility by increasing memory.

You should do whatever suits you.

 

[Searcher214]

Let's say I'm a typical SOHO user with a 100 € budget... I can afford either cache or memory upgrade. I would realize greater utility by increasing memory.

But in this case we already have 32 GB RAM and two 1 TB NVMe’s available for cache.

You should do whatever suits you.

What suits me best is only what will offer best performance, without sacrificing reliability, for usage case described above.

I’m trying to figure out whether the excess RAM justifies setting cache as read only, versus read/write, for the usage case described previously. Can you offer an opinion on that?

 

[Telos]

I’m trying to figure out whether the excess RAM justifies setting cache as read only, versus read/write

Test it with your system, and your usage parameters, and decide. Time to move past the theoretical and speculative.
Post your results here. I've no more to add.

 

[7Natives]

But it's also been stated in both Synology and third-party literature that third-party drives can work when NVMe's are used for cache only. They just aren't testing/listing them, anymore.

To me that makes sense. If they're limited to caching, Synology would have no particular exposure, thus no reason to test; and their Synology Products Compatibility List very specifically cannot list what they have not tested. To do so they would be exposed by endorsing a completely unknown product. The entirety of the issue rests on what support they will or will not render in the case of using unlisted products, and their "limit" in the case of M.2 NVMe's pretty much stops at pulling the M.2 and seeing if the issue continues — unless through some super-weird confluence of events the M.2 fries with a bang and somehow damages the hardware it's attached to.

I'm not saying I don't get the frustration, I do. I've been a third-party HW advocate since forever and still am, but I can see the manufacturer's side in it too and don't think it's unreasonable. When they were sending up ceaseless red panic flags about "INCOMPATIBLE DRIVE!!!" … that was w-a-y over the top! Now that they've reined that in the current approach seems entirely reasonable — to me at least. Others obviously disagree, which is fine and as it should be.

 

[Searcher214]

Test it with your system, and your usage parameters, and decide. Time to move past the theoretical and speculative.
Post your results here. I've no more to add.

Sounds like a plan. The only trouble is, with so many processes / activities running on the NAS, I don't think I can rely on any simple single open/save/close testing. It has to be more statistical testing, based on a large number of ops.

Is there a good benchmarking software that I should be considering for this?

The remaining factor is that my NVMe lack the Power Loss Protection, which may be a slight argument against using it for write cache, even if arranged in RAID-1. I'm honestly not finding any NVMe drive listed as compatible (3rd party sites) with Synology DS and having PLP. It's possible they don't exist.

 

[7Natives]

The remaining factor is that my NVMe lack the Power Loss Protection, which may be a slight argument against using it for write cache, even if arranged in RAID-1. I'm honestly not finding any NVMe drive listed as compatible (3rd party sites) with Synology DS and having PLP. It's possible they don't exist.

A potential cure for the PLP issue might be attaching a pure-sine wave UPS. CyberPower makes units for this, they're pretty decent and fairly priced. The batteries usually last 3- to 5-years and 1-replacement pretty much takes up the useful life of the internal hardware for the unit. Battery life is dependent on your individual utility supply. The reason for the pure-sine wave vs a stepped-sine wave is (imo) really down to the sensitivity of what you're trying to protect, e.g., the NVMe's, etc. The argument for an always-on UPS is simple: If you run a straight battery backup there could be enough of a skip-to-battery-to-utility (remember, milliseconds here) to cause issues with the chips, plus those are all stepped-sine wave.

 

[Searcher214]

I'm honestly not worried about my UPS's. I have a fleet of APC Back-UPS Pro and Pro-S models, that power the NAS's and critical workstations. They perform just fine. The probable source of power failure would be between the UPS and the SSD, such as NAS power supply, cordage, or CPU.

Honestly, I think the only probable cause of issues would be someone knocking that silly PS-2 style connector loose from the back of the NAS. I loathe those wimpy connectors.