Romex Software Forum

Did anyone tried PrimoCache server in combination with a Intel 900P Optane card yet?
We intend to use either 2 or 4 Optane cards as cache for our large server volume. It should fill our 40 GBit Ethernet connection to almost maximum then and reduce reaction times by a factor of 10 over current Samsung 850 EVO cache drives.

If anyone tested that in real life, please let us know!

Due to the fact these cards are so fast, there risk of data loss over SSDs ist about 1/10th too.

Not much of a market for that particular product, despite the marketing on it. Gamers may buy it, but they'll realize little acceleration over current high-end technologies. I can't imagine that putting Primocache on top of it would really offer that much more of an improvement, but I don't have an Optane to test. Haven't ever heard of anyone I know buying one either. Would like to see it, if only just out of curiosity.

@Axel Mertes, actually we're testing the storage performance with Intel 900P Optane cards on server systems. So far we only have 1 card. But later if possible we'll get more cards.

@Jaga @Support
Here is a test of the Optane 900P from toms hardware. This chart pretty much says it all:


taken from toms hardware, see here (chart is from page 3):
http://www.tomshardware.com/reviews/int ... 292-1.html

This is my favorite cite from the test:
"Given the baby steps we see with flash technology, it could take at least twenty years for it to catch up to Optane if it ever does at all."

The main point is that even a PCIe NVMe SSD isn't really faster than a SATA SSD for low queue depth. The Optane is insanely fast and delivers about the same speed at any queue depth. Nearly 2.5 GByte/s when reading small files is more than amazing. Nearly 5 GBytes/s with two cards.

Clearly I see such values on the latest SSDs, but only for streaming big files AND big queue depth. You also have to keep in mind: The Optane answers about 10 times faster than the SSD. So for small files the data might already be received in full before the SSD even begins with the transfer. So these delays add up heavily when comparing both. I expect being able to fill up my 40 GBit ethernet connection on the server with two Optane 900P cards, so we get maximum performance. PrimoCache Server runs since years stable on our Server for read caching and also read/write caching for backup mirrors. I haven't tested version 3 yet on the server, though.

The jump we have seen from using HDD only to HDD plus SSD caching on our windows server shared drives was dramatic. Unfortunately using write caching is felt unsecure. With Optane we might see another jump of the same kind, reducing reaction times my magnitudes again and filling ethernet to the limits. Due to the faster reaction, it might be more safe to use it even for write caching, especially if one implements a mechanism to write back write cache blocks AFTER a reboot/crash/power outage etc. to fulfill whatever possible for a journaled file system to recover. No idea if that is in the list for v3 of PrimoCache...

Here is another interesting comparison picture thats says it all (for my kind of business):


taken from:
https://www.tweaktown.com/reviews/8423/ ... index.html

this got me thinking Primocache might need L3 cache layer for the normal SSD \ NVM and use the L2 for optane memory (consumer 16 /32 version). I was thinking of upgrading my hardware later this year since bunch new hardware will be coming out. So pretty much, using the L3 as spill over from the L2 optane.

Not all consumer can go out and buy the P900 for playing around the latest tech. I think Linus tech also this a review on the P900. On a day to day use, normal usage would not any gain over regular SSD but intent real/write task greatly benefit small files or scratch pad utilization.

As these posts are off the original thread, I move them here and make a new thread for this topic.

I found this article (and the accompanying comment discussion) to be of interest. It highlights that synthetic tests for this card/drive aren't equating to real-world performance, and that the Optane 900P isn't quite the actual boost that everyone wants (or that they are advertising) especially for the gaming segment of the population (those people they are bundling Star Citizen for).

https://www.anandtech.com/show/11953/th ... p-review/2

Of interest are that page, and the Conclusion page. For the pricepoint, it's not quite what I'd call worthwhile, though it *is* a higher performing piece of technology, that can't be disputed.

It will be interesting to see what Support comes back with for tests. I'd like to see some real world tests too, not just sequential/random reads/writes.

support wrote:@Axel Mertes, actually we're testing the storage performance with Intel 900P Optane cards on server systems. So far we only have 1 card. But later if possible we'll get more cards.

@Support:
Hi Support,

can you elaborate on your results so far?
Or maybe simply share some benchmark results like with ATTO disk benchmark or Blackmagic Design Disk Speed Benchmark?
That would shed some light on this,
It would help me in my buying decision.

I know that Intel has its own caching software (CAS), but it has far less options, so I really want to use PrimoCache with this.

Cheers,
Axel

@Axel Mertes, we haven't finished all kind of testing yet. But, yes, we already got some benchmark results. I will upload some results next week.