Hi there,
I'm currently trying to pick an appropriate enterprise SSD for my Proxmox VM set up. I have a consumer SSD and per best practice from the forums here, they don't cut it. And sure enough, after during my daily tasks, IO delays are a problem.
I found this article that recommends a 50-100 IOPS per GB per VM. Now I did some calculations based on manufacturer provided specs which at least for one manufacturer are embellishing the numbers a bit.
316 and 666 IOPs per GB totally can't be true or my VMs would be running extremely smooth and it doesn't. Maybe works totally fine for a consumer PC but for VM it doesn't cut it. Doesn't even cut it for a Proxmox host drive since SWAP is ran off the host drive by default at least. So probably a bit of embellishment of the stats on the manufacturer's part.
This seems a lot more accurate. My host input delay does still struggle a bit with this consumer SSD drive and far short of the 50-100 IOPS per GB per VM recommendation.
I see this used drive on ebay for pretty cheap. The IOPS per GB definitely meets the recommendation from that article.
This particular SSD is also pretty cheap used. Proxmox did a benchmark and this performed much better than consumer SSDs. The read IOPS per GB is pretty high but the write IOPS is low. I guess the read compensates for the write. Maybe not great for constant writing for long periods of time.
Any thoughts on this method of gauging what SSDs to pick? Thanks.
I'm currently trying to pick an appropriate enterprise SSD for my Proxmox VM set up. I have a consumer SSD and per best practice from the forums here, they don't cut it. And sure enough, after during my daily tasks, IO delays are a problem.
I found this article that recommends a 50-100 IOPS per GB per VM. Now I did some calculations based on manufacturer provided specs which at least for one manufacturer are embellishing the numbers a bit.
Inland Professional 120GB | 38,000 IOPS Read 4k | 80,000 IOPS Write 4k
316 IOPS Read Per GB | 666 IOPS Write Per GB316 and 666 IOPs per GB totally can't be true or my VMs would be running extremely smooth and it doesn't. Maybe works totally fine for a consumer PC but for VM it doesn't cut it. Doesn't even cut it for a Proxmox host drive since SWAP is ran off the host drive by default at least. So probably a bit of embellishment of the stats on the manufacturer's part.
Samsung SSD 870 EVO 1TB | 13,000 IOPS Read 4k | 36,000 IOPS Write 4k
13 IOPS Read Per GB | 36 IOPS Write Per GBThis seems a lot more accurate. My host input delay does still struggle a bit with this consumer SSD drive and far short of the 50-100 IOPS per GB per VM recommendation.
Intel DC S3700 200GB 2.5 inch| 75,000 IOPS Read 4k | 32,000 IOPS Write 4k
350 IOPS Read Per GB | 160 IOPS Write Per GBI see this used drive on ebay for pretty cheap. The IOPS per GB definitely meets the recommendation from that article.
Intel DC S3500 120GB 2.5 inch| 75,000 IOPS Read 4k | 4,600 IOPS Write 4k
625 IOPS Read Per GB | 38 IOPS Write Per GBThis particular SSD is also pretty cheap used. Proxmox did a benchmark and this performed much better than consumer SSDs. The read IOPS per GB is pretty high but the write IOPS is low. I guess the read compensates for the write. Maybe not great for constant writing for long periods of time.
Any thoughts on this method of gauging what SSDs to pick? Thanks.