Newbie with 3 questions

[pedz]

Right now, I am trying to store all of my photography and past old data on a single place and I also plan to have some type of cloud backup solution. I currently have 2.5T in Dropbox and I can easily see this getting to 10TB this year because I want to go through and get all the old data I have on countless hard drives that I've kept over the years. For now, I'm assuming 10TB will be a good guess for the near future. The traffic to and from the NAS I believe will be minimal with basically more new data being dumped to it from my laptop -- the data coming from my cameras and seldom reading from the NAS but, obviously, I will read from it on occasion when I go back and look at old work, etc. The laptop I'm getting will have 4TB on it so I'm assuming I can cache stuff up on the local laptop efficiently.

Q1: I plan to buy a TrueNas / iXsystems system. All of my house is 1G ethernet -- which is basically just a router and tons of WiFi. The "plus" systems come with 10GB ethernet which I mostly don't need right now but my real question is: is there a simple way to connect 10G ethernet to a 1G router? The "plus" systems do not cost that much more and offer more versatility which is the reason for the question.

Q2: I will be new to ZFS and my general perhaps extremely vague question is: when my original set of drives start to fill up, is it plausible to replace the smaller drives with larger drives? I assume this won't be a cake walk but I'm unclear if it is even possible. This might be a "depends upon how I set up the NAS" type question so please assume I know nothing and point out the accepted approach to allowing me "grow" the size of the individual HDDs over time.

Q3: The Mini X has 5 3.5" slots and 2 2" slots. I was going to put 4 4TB HDD drives and a 2TB SSD in the 3.5" slots and put the read cache and write cache in the two 2" slots. Does that sound like a decent first start?

 

[Evertb1]

Q1: That should not be a problem as there are normal Rj45 connectors available. But why would you? Al those mini's have 4 1 GbE interfaces available as far as I know (plus the Rj45 for the IPMI).

Q2. Yes that is certainly possible. You will be able to replace disks with bigger ones. The manual has instructions for that. Just be aware that the extra diskspace will only become available when that last disk is replaced.

Q3. For most applications you should not bother with read and write cache. Search the forum on this subject. A ton of information is available.

 

[ConKbot]

Q1: I plan to buy a TrueNas / iXsystems system. All of my house is 1G ethernet -- which is basically just a router and tons of WiFi. The "plus" systems come with 10GB ethernet which I mostly don't need right now but my real question is: is there a simple way to connect 10G ethernet to a 1G router? The "plus" systems do not cost that much more and offer more versatility which is the reason for the question.

They come with ethernet 10gig ports, so you can just plug in a normal ethernet cable (cat 5e is fine if thats what you have around, no special requirements at all), into your normal 1gig router/switches and get a 1gig link. Easy as can be.

Q2: I will be new to ZFS and my general perhaps extremely vague question is: when my original set of drives start to fill up, is it plausible to replace the smaller drives with larger drives? I assume this won't be a cake walk but I'm unclear if it is even possible. This might be a "depends upon how I set up the NAS" type question so please assume I know nothing and point out the accepted approach to allowing me "grow" the size of the individual HDDs over time.

A vdev can be expanded by replacing the drives one at a time, and resilvering after swapping each disk. Once the last one is replaced, the pool can grow. The specifics of this can be expanded on by others, as I haven't actually done this myself.

Q3: The Mini X has 5 3.5" slots and 2 2" slots. I was going to put 4 4TB HDD drives and a 2TB SSD in the 3.5" slots and put the read cache and write cache in the two 2" slots. Does that sound like a decent first start?

For your application, neither of those would be necessary or useful.
OpenZFS (which TrueNAS uses) system RAM a the disk cache. A system will use pretty much all of the free space in ram for caching. The read cache (ARC) of the most recently and most frequently used data (since boot, it does not restore the cache after a restart) . The write cache is 5 second long blocks of asynchronous write data to be written to the array, called transaction groups

L2ARC (the SSD 'read cache') holds data that gets removed from the ARC with other data being deemed as more important. But the RAM has to hold a table of contents of the L2ARC. Reducing the available amount of (extremely fast) memory available for ARC duty. Given it sounds like youre looking at 1 (maybe 2?) users, and just a network share, nothing intensive, you 100% are fine without an L2ARC. Not fine as in "you dont need the extra performance," but fine as in "it will do nothing for you" Given your limited amounts of read you say you will be doing, the L2ARC may not even start to be used.

SLOG (sounds like a write cache, but it is absolutely not a write cache) is only ever used for synchronous writes. i.e. network protocols that demand the data in transfer be stored onto non-volatile memory before the write is marked complete. For a windows network share (i.e. SMB/Samba) sync writes are generally not used, so this drive would sit in your server, unused.

To expand on what the SLOG does so you can understand it better, for sync writes, data comes into the NAS, and written to the SLOG immediately, along with being put into the RAM for the 5-second transaction group. Once the data is on the SSD, the server can reply back to the client that the write is complete. In normal operation, the system closes out the 5-second transaction group, writes it to the array, and then that space on the SLOG can be freed up. If you have a power-fail during that 5-seconds, upon next boot, the NAS looks at the SLOG, takes the data that was stored there, and writes it to the array. This is the only time the SLOG is read from, and why it is not a cache. If you need a SLOG in a system, you dont want just any SSD, as the SSD cache has to be bypassed for the sync writes (or the cache written to and purged with every write) making effectively a cacheless SSD, and those perform like garbage. A SLOG SSD should be one with data-in-flight power-loss-protection, so that it can use the DRAM cache on the SSD. The data gets written to the SSD DRAM cache, the write is marked as complete, and if power fails, the SSD onboard capacitors have enough power for the disk to purge the cache to flash memory, so the data is safe once its in the SSD DRAM, and you then have the "full" SSD performance.

And besides, if those are cache disks, where are you going to put boot disks ;)

For a bit of further assurance to you, that you're fine without the SSD caches (I know, they sound like a really good idea on the surface level) I run a NAS with 6x 14 TB drives in a RaidZ2 array. 84 TB raw capacity, 128 GB of ram, so ~1.5GB ram per TB of raw disk capacity. If you put 5x 4TB disks into the mini+ in a RaidZ2 array, with the base 32 GB of ram, for 10-12TB of formatted capacity, and 20TB raw capacity, you would be looking at 1.6 GB of ram per TB of raw disk space. Plenty of room for the ARC to do its thing for both of us. RAID Z arrays will be limited to the random IOPS of a single disk, but in a network share type of application like this, you rarely are doing truely random IO. If I restart my server so the ARC is completely empty. Then start a copy of a ~3-4 GB folder of ~1.5MB images, I still get full ethernet line rate (925+ mbps) for the transfer off the NAS. If you were storing a ton of KB size files there may be a slowdown in that sort of situation, but given you say you are a photographer storing your files. I doubt much of your data is KB size files?