I'm in the process of deploying a new two-node VM system based on Proxmox to replace our vSphere setup, and I've hit a snag with Starwind performance. My hardware includes two Lenovo SR630v3 servers, each equipped with a Xeon Silver 4514Y CPU, 64GB of RAM, and an array of NVMe and SATA SSDs managed by different RAID controllers. I've set everything up with a 10G network and have tested various configurations, including MTU sizes of 9000 and 1500 with little to no performance difference.
The main issue arises when I create a LUN in Starwind and connect it to Proxmox. Without write-back caching enabled, the write speed is painfully slow at around 2MB/s; with caching on, it jumps above 100MB/s. However, I'm worried there's a deeper issue at play since write-back caching shouldn't be necessary to achieve decent performance.
Currently, I'm using a RAID 5 array created in the host UEFI for the storage pool that Starwind recognizes. I've toggled between default and custom configurations including ZFS but haven't noticed any significant performance gains. It seems like I can't access certain options like "hardware RAID" either.
In contrast, when I create an LVM on the RAID drive and bypass Starwind, I get write speeds of 200-300MB/s. I've done some iperf tests, and the internal network performance looks solid at around 9.8GB/s, so it feels like the issue is isolated to Starwind's configuration or performance.
Does anyone have suggestions or insight into what could be causing these slow speeds?
2 Answers
You might want to check the CPU settings for your Starwind VM. Make sure the CPU type is set to 'Host' for better performance. Also, if you're experiencing low speeds, remember that without write-back caching, both nodes must first write the data to disk before acknowledging the write, which can really slow things down based on your setup. If you haven't already, try passing the drives directly to the Proxmox VM instead of using a single volume from the RAID controller—I've seen better performance that way. Any support from Starwind you could tap into might help too!
Thanks for sticking with Starwind! CPU emulation can indeed impact performance, so definitely verify that your CPU type is set to 'Host'. I recommend opening a support ticket so we can do a more detailed configuration review. There are many factors at play like RAID settings and multipathing that we're unable to assess thoroughly through comments alone, and we'll update the community on your case once it's resolved. Don't hesitate to reach out if you have any more questions!
Thanks for that insight! It sounds like the sync delay might be the culprit. I did try setting up a standalone LUN with no replication but didn’t see much difference either. Unfortunately, I don’t have support at the moment since I’m using the free version.