Romex Software Forum

Im new to Primocache. I purchased a server and listened to the rep who sold me on a RAID card that did not support caching. I know, my fault for not researching.

So, slow writes are a problem.

My setup is Windows Server 2016 Essentials running on a dual processor with 32gb of RAM and 4 2TB mechanical HDDs in RAID 10.

Looking to hear from experienced folks on suggested settings.

Currently have:
OS managed memory: 12928mb
Block size: 64kb
Defer write: 20s
L2 Cache on a 264gb USB set to MAX
No Invisible memory.

Defer-Write will greatly improve write performance. However, a power outage or system failure (crash/freeze) might result in data loss or corruption because in such scenarios the cache has no chance to write data back to the disk. So it is recommended that you may only enable Defer-Write on volumes where temporary, unimportant or reproducible data is to be stored.

If you have to enable Defer-Write, you may use a small latency like 10 seconds or smaller. Smaller latency suggests that less amount of deferred write-data is in the cache which might get loss on a failure.

On a server system, it is not recommended using a USB drive as level-2 cache. SATA/NVMe interface is much better than the USB interface.

support wrote: ↑Mon Dec 23, 2019 4:04 amSo it is recommended that you may only enable Defer-Write on volumes where temporary, unimportant or reproducible data is to be stored.

Or, when using a UPS to protect the system in question, so you can achieve graceful shutdowns/hibernations. That's what I do, and I haven't had a problem with the deferred write cache in a very long time.

Jaga wrote: ↑Tue Dec 24, 2019 5:11 am

support wrote: ↑Mon Dec 23, 2019 4:04 amSo it is recommended that you may only enable Defer-Write on volumes where temporary, unimportant or reproducible data is to be stored.

Or, when using a UPS to protect the system in question, so you can achieve graceful shutdowns/hibernations. That's what I do, and I haven't had a problem with the deferred write cache in a very long time.

How much time do you set defer write?

Yes, I have a UPS in place.

Thanks

steve841 wrote: ↑Tue Dec 24, 2019 7:34 am

Jaga wrote: ↑Tue Dec 24, 2019 5:11 amOr, when using a UPS to protect the system in question, so you can achieve graceful shutdowns/hibernations. That's what I do, and I haven't had a problem with the deferred write cache in a very long time.

How much time do you set defer write?

Yes, I have a UPS in place.

Thanks

Typically about 60 seconds. It depends on your level of confidence for the stability of the machine. The more stable it is, the longer you can set the interval. I've gone as high as 10 minutes, but I don't see much advantage in anything over 60 seconds.

I usually gauge what I'm writing to the disk and do interval of 5, 10, 15, 30, and 60 to see when excess write being force to write to the disk. Took an average and set my deferred time after that. It seem for normal desktop deferred time for me is 10 (default setting) is good enough.

Server workload could be difference depending on critical data and performance required. Most raid controllers set their at 4 - 5 seconds.

minhgi wrote: ↑Tue Dec 24, 2019 5:47 pmServer workload could be difference depending on critical data and performance required. Most raid controllers set their at 4 - 5 seconds.

That is certainly possible, but you're not going to get a lot of trimmed blocks that way, especially on a server. I don't think I'd ever go under 15 seconds, just on principle. But then I've never run Primocache tests on a 2016 server with multiple clients attached. Fortunately it'll be easy for Steve to do so and get immediate results.

Thanks for the advice.

Im testing 15 and 20 seconds for defer write.

steve841 wrote: ↑Tue Dec 24, 2019 6:40 pm Thanks for the advice.

Im testing 15 and 20 seconds for defer write.

Might want to try 15, 30, and 60 for comparisons. A difference of a mere 5 seconds may not be able to give you a decent delta for results.

If you have UPS you can use write strategy "Idle Only" + latency: infinite. This will write data from cache to HDD only when system is idle (or cache is full), so you will not experience write speed drop when cache start flushing to hdd after 5-20-60 second.


Question to @support, will you implement strategy that L2 is "content aware" so if L2 flushed 50% of cached data to HDD and experience power loss, then it will not lose that 50% data remaining on L2, but rather flush it after reboot?