Romex Software Forum

1) Does PrimoCache support Windows Server 'core' versions (without a gui)? My testing shows the CLI exe works, but why doesn't the GUI version at least run on server core? Other GUI apps can run like task manager, etc. Is there some windows feature the GUI needs to run correctly?

2) Also, has the issue been fixed of being able to manage L2 cache using the CLI?

3) Does PrimoCache run as a service? Or must a user be logged in for the application to function correctly?

4) Looking at the change logs it seems as though you release an update pretty much every December. Are you on track for a 12-2020 release? Any significant features making it into this release?

5) When running PrimoCache, if I have 2 separate windows volumes (E: drive and F: drive) with both drives being 60TB spinning disk RAID volumes. Will a single L2 cache do write caching for both of those volumes? Or do I need to create 2 separate L2 cache volumes? (1 for each drive)

6) If I do need to create 2 separate L2 cache volumes, does the 2TB limit for L2 cache apply across the aggregate of both volumes? (IE, is it 2TB total, or 2TB per L2 cache volume?)

I'm trying to determine if the following use case would benefit from PrimoCache.
I have a large video archive server.
Currently 5x 16TB SAS RAID5
Once the 'cache' contained on the hard drives fills up (512mb x 5) my writes will go from 300MB/sec down to 30MB/sec or lower. This only occurs after sustained sequential writes of more than 3GB. Thus this edge case doesn't really show up in standard IO meter or Atto benchmark tests since they are typically just measuring pure IOPS or short bursts of read/write.

My issue is when I am copying large video files to the server (single video files can be 40GB+) I quickly saturate the disk array.

I originally was going to replace my Dell PERC H710P with an LSI 9361-8i and purchase the CacheCade 2.0 hardware key. However after learning about PrimoCache I'm not so sure now. Reading up on CacheCade it seems it works great for random read/write workloads like a database server, but maybe not so much when doing large sustained sequential write operations.

Will PrimoCache help with this? If I were to configure the server with the following ...
L1 Cache (ram) of 4GB
L2 Cache (ssd) of 2TB
HDD RAID5 array

In this scenario, when I would write large 40+GB video files to the array it would be going 100% to the SSD first, then slowing bleeding over to the HDD's (based on various configuration scenarios)
Is this correct? would PrimoCache help with sustained write operations in excess of 40+GB and give me the kind of 300MB/sec performance I'm looking for during the entire write operation? (not just the first few GB like I have currently)

cyr0nk0r wrote: ↑Tue Oct 06, 2020 4:26 am Does PrimoCache support Windows Server 'core' versions (without a gui)?
Also, has the issue been fixed of being able to manage L2 cache using the CLI?

Yes.

cyr0nk0r wrote: ↑Tue Oct 06, 2020 4:26 am Does PrimoCache run as a service? Or must a user be logged in for the application to function correctly?

It's running at kernel level. No login is required.

cyr0nk0r wrote: ↑Tue Oct 06, 2020 4:26 am In this scenario, when I would write large 40+GB video files to the array it would be going 100% to the SSD first, then slowing bleeding over to the HDD's (based on various configuration scenarios)
Is this correct? would PrimoCache help with sustained write operations in excess of 40+GB and give me the kind of 300MB/sec performance I'm looking for during the entire write operation? (not just the first few GB like I have currently)

Yes. With Defer-Write, your writing performance should be improved. However, please note that a power outage or system failure might result in data loss or corruption because in such scenarios the cache has no chance to write data back to the disk.

Thank you. Good to know PrimoCache will support my use case.
Any word on the remaining questions in the original post?
Specifically, 4, 5, and 6 (and the 2nd part of question 1)

Also, another question.
I assume, but I can't find anywhere where it's explicitly stated.. that data coming in (write cache) will also first go to L1 cache until its write percentage is full. Then it will immediately begin flushing write caching from L1 cache to L2 cache (provided L2 cache is configured and set up for write caching). Then after configurable options/intervals the write cache will flush from L2 cache down to disk.

Do I have this correct? If this is correct, if your L1 cache is say 4GB, and your L2 cache is say 16GB what happens for file transfers of 40-50GB sequential write? Is the data constantly being moved from L1 to L2 to disk as fast as its coming in? once both L1 and L2 cache is full and data is STILL coming in what happens?

cyr0nk0r wrote: ↑Tue Oct 06, 2020 4:26 am why doesn't the GUI version at least run on server core?

PrimoCache GUI program needs some library files which the server core doesn't install.

cyr0nk0r wrote: ↑Tue Oct 06, 2020 4:26 am 4) Looking at the change logs it seems as though you release an update pretty much every December. Are you on track for a 12-2020 release? Any significant features making it into this release?

We are still improving PrimoCache. See viewtopic.php?p=14983#p14983

cyr0nk0r wrote: ↑Tue Oct 06, 2020 4:26 am 5) When running PrimoCache, if I have 2 separate windows volumes (E: drive and F: drive) with both drives being 60TB spinning disk RAID volumes. Will a single L2 cache do write caching for both of those volumes? Or do I need to create 2 separate L2 cache volumes? (1 for each drive)

PrimoCache supports one cache for multiple target volumes.

cyr0nk0r wrote: ↑Tue Oct 06, 2020 4:26 am 6) If I do need to create 2 separate L2 cache volumes, does the 2TB limit for L2 cache apply across the aggregate of both volumes? (IE, is it 2TB total, or 2TB per L2 cache volume?)

2TB per cache task.

cyr0nk0r wrote: ↑Tue Oct 06, 2020 5:33 pm that data coming in (write cache) will also first go to L1 cache until its write percentage is full. Then it will immediately begin flushing write caching from L1 cache to L2 cache (provided L2 cache is configured and set up for write caching). Then after configurable options/intervals the write cache will flush from L2 cache down to disk.

In 3.x, when L1 is full, it will flush to target volumes. in 4.x, there is an option allowing to flush L1 to L2 when L1 is full. If both are full, "urgent write" will raise and flush certain data to target volumes even the interval time not expires.

PS. Please don't directly change the original post, just add new content to the end of the original post and marked as EDIT. Thanks.

Support wrote: ↑Wed Oct 07, 2020 4:59 am In 3.x, when L1 is full, it will flush to target volumes. in 4.x, there is an option allowing to flush L1 to L2 when L1 is full. If both are full, "urgent write" will raise and flush certain data to target volumes even the interval time not expires.

Hmm, this is a pretty significant statement.
So in 3.x if L1 is full it will flush to spinning disk volume and not go to L2 cache? If that is the case, what is the point of write deferred L2 cache if the data never makes it into L2 cache?

It sounds like in order to get write deferred data into L2 cache you'd need to either
a) not have an L1 cache at all ... or
b) configure L1 cache to be 100% read

Is this correct? If so, what is the recommendation to get data into write deferred L2 cache?

When L1 is full, L1 starts flushing certain amount of deferred write-data, while L2 starts caching incoming write-data.

Support wrote: ↑Wed Oct 07, 2020 5:25 am When L1 is full, L1 starts flushing certain amount of deferred write-data, while L2 starts caching incoming write-data.

I don't understand how this is any different than "when data comes in it first goes to L1 write cache until full then goes to L2 write cache"
How is your statement different than mine?

EDIT: Ok, I think maybe I understand what you're saying.
It sounds like what you're saying is that when data is coming in at a constant write (enough to fill up L1 cache) that once L1 cache becomes full it will start to flush the data from L1 cache to the spinning disks. At the same time, the incoming data is no longer coming in to L1 cache first, it's now coming in to L2 cache since L1 cache is busy being (a) full, and b) flushing its data to disk). At this point as the data continues to come in since L1 cache is full, the L2 cache takes over as the write deferred cache. Do I have that right?

If so, once L1 cache has finished flushing its data to disk, does it take back over and start accepting new incoming data again?

Yes, exactly.

Can Primo Cache not have an L1 cache at all?
rxpcc.exe new -v E,F -m 2048 -l <guid> -r 90,90 -b 512 -w 120 -a average
If I specify -m 0 ... or leave -m out completely will that work?
In a use case of someone not wanting an L1 cache at all, what is the recommendation?
Or would it be better to specify -m 1024 , but then set -r 0,90 ? Thus making L1 cache read cache only?

PrimoCache supports only L2 cache, without L1 cache. So, yes.