ATTO Benchmarks for a 4-HDD Raid0 pool

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

Dreece

Active Member
Jan 22, 2019
503
160
43
The spec:
4 x 12tb SAS Exos12
HPE P840 4gb Raid Controller

State:
All caching disabled, physical and controller.
Pool is RAID0 with a strip size of 256k (1MB stripe).
OS is Windows 2019 Server.
Filesytem is ReFS at 64kb.


test1.jpg

Not bad raw-uncached throughput for just 4 drives, this pool is entirely for large (2gb+) raw video footage archiving.
Once I enable caching plus Smartcache (2 SAS SSD RAID0), my use-case kicks into well and truly served mode, ie. raw footage scrubbing in NLE timeline at virtually no-latency. The old smartcache/cachecade tech still has a use-case indeed.


test2.jpg

Once you enable all the caching, things go insane, though I wouldn't read into the above as anything substantial, it just gives an idea of what happens when working on raw cached data. The server itself has a huge chunk of free ram for OS to cache files too (Windows Server is pretty good for that), along with the 100G RDMA link to client, things fly no different to as if you were working off a local NVME drive and with very little to negligible effect on cpu cycles. Indeed one can achieve greater throughput with more elaborate costly configurations, but for the cost of a night out on the town of fine-dining-and-wining (pre-lockdown era), these results more than serve this old man's use-case.

Backup wise, all data is kept in sync on a different rack server with a non-raid controller and a bunch of HDDs' under ZFS on a 3 day sync (other smaller pools are SSD RAID10 with daily backups), along with a monthly LTO7 archive job.

Once you have the tape setup, you kind of ignore all the worries of data-loss and play with fire like you have superpowers. Obviously this set-up isn't ideal for zero-downtime environments, but for many homelab'ers where a fat bank-balance is a priority and a bonus with the missis, it works, great at that too.
 
  • Like
Reactions: Lix

Dreece

Active Member
Jan 22, 2019
503
160
43
Here's a comparison between the above pool and a PM1725a nvme card, I wouldn't read into these benchmarks as gold for they were run on an active server that does a hell of a lot more than simply serve data (a fair few VMs including 24/7 CCTV capture and motion-detection of a small army of cameras), but for I it is never about being a drag-queen, it's about consolidation and reducing-wattage where possible, and I do work this server hard like a tractor:

Terminals_wxP08PM3iX.png

All down to use case(s) really. The nvme I use for database work, the disk pool I use for serving huge sequential 4K video footage to an editor across the network, which I've layered to 9 levels (ie 9 different 4k 100-200mbit video streams, with cuts/edits back and forth, rarely do I use more than a handful on layers to be fair, I did it just to make certain it can cope) and all smooth as butter, can't argue really.

I was going to stick with a sas3516 chipset card originally but without smartcache it was practically useless for what I was after, a quick and easy caching big-data storage config that just works with a few point and clicks.

Hope all the above motivates the budget-conscious to realise that one doesn't always need to spend silly money on the latest and greatest tech, old-school can work if you configure it just right, and hopefully this then leaves one a greater budget for a more potent CPU and fast tight-cas RAM for the creative workstation.
 
  • Like
Reactions: lostmind