Synology SSD Cache Setup and Testing
HTML-код
- Опубликовано: 11 июл 2024
- How to guide for enabling Synology SSD cache on your Synology NAS. IOMeter demo for testing SSD cache performance and IOPS results of my SSD cache configuration.
TechThoughts corresponding blog article:
www.techthoughts.info/synolog...
Tobu & Syndec - Dusk [NCS Release] • Video RUclips / tobuofficial 
Наука
You need a boatload of cache to make it worthwhile, and disabling sequential access makes it even harder to benefit from cache (because when you copy something in, it doesn't pre-populate the cache with your latest [and probably most used] files).
On our mainframe disk I/O subsystem, there was a bit you could set for "this is a backup" which would prevent backups from poisoning your cache. This make a lot more sense than disabling sequential access (but of course since virtually nothing is vertically integrated in distributed systems, it would be a ginormous task to get everyone to agree to changes like this).
And this is why I plan to go to an all SSD NAS. Never having to worry as much when it comes to these types of situations, and even when you do use M.2 for cache you see huge leaps in file response.
I noted that you did not put the SSDs in the correct slots. The SATA backplane in the Synology 1815+ is configured so that Slot 1 & Slot 2 are on one channel, and slots 3-8 are on another channel.
Synology uses bcache for handling the tiered caching of btrfs/shr. Synology has configured bcache (IIRC) so that it considers data to not be written until the data is written to both the cache layer and the backing storage. This is a write-through cache. You can actually change this with a little fiddling. Synology's default configuration is to protect your data at all times so it doesn't cut corners with bcache.
When writing data, the synology writes the data to the cache (write operation), and then bcache reads the data from the cache (read operation), and writes it to the backing store (write operation). Again, to re-iterate, data is not considered saved until the backing storage confirms that the data has been safely written there.
When reading data, the Synology first looks in the cache, and if it isn't there, it loads from the backing store (read operation), puts the data in the read cache (write operation), then reads the data from the cache (read operation).
This is why your IOPs are in the toilet. bcache can give an immense performance boost, but unfortunately the way it is configured in the Synology, and rightfully so if you want your data to be as safe as it can be, is to actually introduce a performance hit until the cache warms up.
You can modify your Synology configuration -- tweaking bcache to use write-back caching -- so that data is written to the cache directly, and then written to the backing store separately, and that will speed up your write IOPs. But there is a chance, in the event of cache failure or cache corruption, that you will lose data. You can also change the configuration so that if data is not in the cache, it is delivered directly from the backing store first, and only then is it transferred to the cache.
It isn't anything to do with the h/w in your configuration, it is all to do with the way Synology configures bcache to keep data as safe as possible.
P.S. I may be a little hazy on the bcache read/write operations, it has been two years since I looked at the source code & documentation. I do believe that bcache writes then reads then writes, the read operation being to check that the data written to cache matches the data that was meant to be written.
P.P.S. Get some cheap Enterprise SSDs from eBay and toss out those junker consumer SSDs when building cache drives. You will also then not care about the "don't use cache for sequential I/O setting" as the write characteristics for the TDW will be measured in several petabytes rather than a couple of hundred of terabytes and your IOPs ability will also be far higher.
P.P.P.S. I would double-check my statement on the write-back vs write-through caching as it has been a while since I fiddled with this stuff and Synology may have updated their caching configuration.
Thanks for making this interesting and informative video on SSD caching. I am currently using a 2-bay Synology and considering of upgrading it as DS420J, do you think I could config something like 2x8TB HDD (RAID 1), and 2x128GB SSD (RAID 1) as cache?
do i need to enable the cache inside a 2 bay DS718+ with 2 ssd of 4TB ? (no magnetic hard drive here only those 2 ssd) thank you
Hello, I have DS1515 +. In the last 2 positions I have installed 2x SSD Crucial MX300 525GB. I do not know what it is, but I have not experienced any higher system performance. On the contrary, copying larger files is slower ... SSDs are in my case in positions 4 and 5.
Is the write cache ONLY for files that were already written to the SSDs? What about file transfers to NAS from your PC - write cache actually help on the initial / first file transfer?
BEST TO USE SLOT 1 AND 2 FOR SSD CACHE PER SYNOLOGY DOCUMENTATION
Can you help me a bit?
I'm not sure if I will do the right thing here, I have two 240GB SSD's and I want to insert them in my HP Microserver Gen8.
If I will do that, I will have to settle for 2TB of data support, because I have to keep the redundancy in place, so I will lose the other two slot for that option.
I use this unit only use Xpenology for a PLEX server, so I need to know if those SSD's will really help me. All I want is to watch any kind of 4K with no buffering or interruptions...
Lately I saw that the large 4K files doesn't move very well in my actual setup, despite the fact that I have a Xeon microprocessor and 8GB RAM.
Great video! I was wondering how we should calculate cache size vs overall size of installed drives.
Synolgoy DSM has an "SSD Advisor" built in to the software that can estimate what you capacity SSDs you will need.
All my bays are full, can I use there cache with an expansion module ?
Per synology forums, The ssds for cashing should be installed in bay 1and2. You should try it out and report testing.
Good call. I had originally stuck the SSDs in slots 7 and 8. I moved the SSDs to slots 1 and 2 today and re-did the testing. Results are the same. Cached data slowly gets amazing but non-cached share files suffer a significant performance penalty when the cache is turned on.
Thanks, so to install the SSD onto the bays you just need to use the supplied screws that came with the SSD's? I know for the regular NAS drives you just snap them in, I was just wondering if the SSDs requires a bracket.
For my DS918+ they required no tools, just slide in
Nice video! I have 2 SSD's I could use, but they are different in size. Do you know if they have to be equal GB's?
Did you try it?
Are these tests sync or async?
How is the NAS connected to your computer? 10gbe or 1gbe? I'm only getting 106mbs on my 1517+ to iMac. Might have to use SSD cache.
thats propably due to your 1gb port on the iMac. 106mbs sounds about right for that
300+ IOPS does not mean 300MB/s . And regarding heavy IO you wont get 106MB/s on the 1517+ . In the video you see there are 300+ IOPS @ 1.24 MB/s at min 8:17
I'm amazed it got 200 views. What a horrible video. Good effort, but, shouldn't you KNOW when you don't know something? Don't you know enough about computers, to know you're a novice?
Why wasn't RAID0 possible to choose for the SSD combination of 2 of them?
Wouldn’t it be better just to use cache for « read » while making initial configuration?
In this case it will not affect the write speed
Yeah, because in the case of NAS systems your benefit is on the READ side of things which is where you want the performance, the added benefit is you'll also get fast Write speeds also.
adding flash memory to anything has an initial hit to performance. Same thing is seen on SSHDs.
Once the flash memory learns of the most accessed files, its faster than average.
Got it, and that is happening here. However, my major concern is that non-cached data seems to suffer a 30% penalty when the cache is set up. I don't know if a 30% drop for all non-cached data is worth amazing performance for the cached data.
So, what did you guys conclude? I'm picking up a NAS today DS918+ and I don't know if I should get 2x 128 GB SSDs or not as I hear conflicting reports over the utility of using them. Heard that it'll load Plex metadata faster and extend the lifespan of HDDs. But if non-cached data suffers, I don't know.
Fennec same situation, so did you install cache at the end ?
Hi, I have 4 of those Synology 1815+ and I made the same test as you did a year ago. But I was not that patient and quickly dropped the idea of caching my Synology.
But I read somewhere on the web some interesting stuff regarding the 1815+, (cannot remember where) And they said that the 1815+ is constructed using Sata 3, Sata 2 and Sata 1.
Bay 1, 2 is Sata 3
Bay 3, 4, 5, 6 is Sata 2
Bay 7, 8 is Sata 1
And this might explain your poor SSD cache performance. Try moving the SSD to Bay 1, 2 instead.
I moved the SSDs to slots 1 and 2 today and re-did the testing. Results are the same. Cached data slowly gets amazing but non-cached share files suffer a significant performance penalty when the cache is turned on.
Does the larger amount of storage SSD make a difference?
I have 1TB read-only cache and it makes things worse because it consumes lots of RAM :))
Dare to dream: would be nice if Synology gave you the ability to turn off SSD caching at the share level. I don’t need non-production data to use the cache, even though some of that gets accessed a lot (thereby being moved to the cache, taking up that cache space that other prod data could use.)
I think you can use two Volumes, one use cache SSD and fast hard drives. The other doesn't.
Do i have to power off my synology to plug ssds into it?
sure lol. Shut it down.
Do you have the same test for freenas/zfs?
Remember no all NAS systems can take advantage of the Hard Drives you use depending on its Transfer Speed Limits and depending on how good you Internet and Network Speeds are.
Hmmmmm, After all the data is cached (all in SSD), 3x read/write ; but before all the data is cached (some in HHD & SSD),
Stick with the read cache...
You need to Get a 10Gbe connection. Caching is useless when you’ll never exceed 12bps
12bps lol is it loading off of punch card?
He is measuring Iops though not sequential reads. 100-300 Iops test is not going to line rate a 1GBe.