Hello friends!
Lemmy.ca will be undergoing scheduled maintenance from approximately 12:00 - 12:30 PT (3:00 - 3:30 ET), while we move to a more powerful server.
Happy Canada day!
Hello friends!
Lemmy.ca will be undergoing scheduled maintenance from approximately 12:00 - 12:30 PT (3:00 - 3:30 ET), while we move to a more powerful server.
Happy Canada day!
We won’t accept anything less than extremely fast!
It’s a big difference going from a 8vCPU 8GB RAM VPS where the actual resources were likely quite over subscribed to a dedicated server with a 6c/12t Ruby Lake Xeon, 32GB RAM, 2x 512GB NVMe drives in RAID with it’s own 1Gbps Public connection. From the data we have it should be sustainable with this server until about 40-50K users.
We’re currently paying ~$140/month for that, and have plans in place to scale up as needed when it’s needed. Over the next few days we will reopen the ability to make donations to cover the costs of operations.
Was it CPU or memory bound prior to the upgrade? Or IO?
All of the above, but more so memory and IO.
Disk or network IO?
Disk, due to low memory. Not enough to keep enough of the db in memory, and having to hit disk for pictures as well.
Oh I see. Makes sense. Do you know if the storage was NVMe previously or spinning media?
It was a VPS and I think on some sort of shared SATA/SAS SSD array, just going off the 300-800MB/s reads I was seeing.
Old box: Timing buffered disk reads: 2066 MB in 3.00 seconds = 687.62 MB/sec
New box: Timing buffered disk reads: 1022 MB in 0.31 seconds = 3338.77 MB/sec
Nice uplift. Random IO which is likely what the db does are probably through the roof compared to the sequential uplift.
The DB is small enough it’s all in ram now, seeing a 100% cache hit rate in postgres.
@[email protected] > Over the next few days we will reopen the ability to make donations to cover the costs of operations.
I’ll be present :)