VPS Plex Server with Cache not performing well

Every now and then it drops to a few kilobytes for a matter or seconds and then jumps back up

I do see this error at the beginning of the log

2018/04/11 12:20:20 ERROR : plex: websocket.Dial ws://localhost:32400/:/websockets/notifications?X-Plex-Token=g2n8DbVdtVdsUxoQDHif: dial tcp [::1]:32400: connect: connection refused

I completely believe there could be intermittent networking issues. But here is some network performance while doing a copy from the rclone mount to a local directory. With averages well over 100mbit, I’m not sure why plex seems to struggle so much. For the time being I have removed the plex integration with rclone. Because for all activities outside of playback it limits the number of workers to 1.

Here is a screenshot a bit later during the same copy and it looks like there are times were there is no network traffic for 10-15 seconds.

Doesn’t that risk too many API hits (during scans) and getting a 24Hour ban?

How do you get that graphical output when copying?

I’m using nload to monitor network traffic, that is what i provided the graphical output.

I’ve rarely see above 50,000 api requests in a 24 hour period. My understanding of the workers concept with plex integration, is that it allows rclone to dedicate workers to playback, rather then routine tasks like scanning or copying files. I posted in a different thread about why rclone copy seemed so fast but when copying using rsync from the rclone mount to the file system, it seemed so slow. Now looking back at my post, the plex integration might be the reason you are only seeing 20-60Mbps during a copy operation.

I removed Plex integration and I am still seeing the network traffic stop to and from the VPS. I’m pretty certain there is something wrong with their equipment.

Let me know what SSDnode’s says. Because I’ve been seeing this behavior for about 2 months Sometimes I can restart the rclone service, plex, radarr, and sickrage and some days everything will behave other days not so much.

will do. I complained to them in February and this was there response

Hello,

We are currently experiencing Virtuozzo bugs on our KVM platform and are working with Virtruozzo to get it resolved. If you are experiencing issues with the network or managing your server in the dashboard, it is due to these bugs. We apologize for any inconvenience in the meantime.

I then complained again in March and they said

Thank you for writing in.

It looks like the server got throttled due to high resource usage. Can you please adjust the software settings so that it does not use high resources for a prolonged time.

Here is our resource usage FAQ.
https://www.ssdnodes.com/manage/knowledgebase.php?action=displayarticle&id=13

I have removed the throttling and rebooted your server. Can you check again now?

I responded that there is no way that I should be throttled as all I am running is a Plex server and i’m not even using it as a primary as the performance is too poor. They said that they had unthrottled it and now it was speed testing as follows

root@ssdnodes-server1:~# speedtest-cli --server 8706
Retrieving speedtest.net configuration…

Retrieving speedtest.net server list…
Selecting best server based on ping…
Hosted by Cable One, Inc. (Dallas, TX) [1953.54 km]: 6.798 ms
Testing download speed…
Download: 1555.80 Mbit/s
Testing upload speed…
Upload: 656.64 Mbit/s

I have never been able to achieve those results when I test it.

SSDnodes response to my support ticket where I asked them to check on network connectivity and relayed my issues with poor upload performance.

"Hello,

The speed test result depends on a number of factors including the test server that is selected and its location.

Here is the result of 2 tests that I ran almost at the same time and the result shows a lot of change. When we select a server without specifying the location, it selects a server in California whereas your server is in Dallas. Please note the result of the second test that clearly shows the difference.

root@ssdnodes-server1:~# speedtest-cli --server 8706
Retrieving speedtest.net configuration…
Testing from Strasmore (104.225.218.00)…
Retrieving speedtest.net server list…
Selecting best server based on ping…
Hosted by Cable One, Inc. (Dallas, TX) [1953.54 km]: 50.37 ms
Testing download speed…
Download: 65.59 Mbit/s
Testing upload speed…
Upload: 122.88 Mbit/s
root@ssdnodes-server1:~#

root@ssdnodes-server1:~# speedtest-cli --server 11207
Retrieving speedtest.net configuration…
Testing from Strasmore (104.225.218.00)…
Retrieving speedtest.net server list…
Selecting best server based on ping…
Hosted by Sprint (Dallas, TX) [1953.54 km]: 13.984 ms
Testing download speed…
Download: 1044.49 Mbit/s
Testing upload speed…
Upload: 360.19 Mbit/s
root@ssdnodes-server1:~#"

Basically blowing me off

I’m noticing the same symptoms but on a real-Hetzner instance when I switched from plexdrive to rclone-cache.

Sometimes pressing play takes a while, it is prone to stutter more, and even the screen where you press play on can at times timeout / need to retry (which i just find strange)

I ended up emailing Matt, the CEO as I have always found him excellent to deal with.

His response is that it could still be the effects of the Spectre and Meltdown patches and then Virtuozzo playing catchup with their software to deal with the knock on performance issues.

His email below.

Hi Jude,

Thank you for reaching out to me via email. Sorry to hear you’re experiencing issues with your server.

This may be still related to some of the Virtuozzo bugs we’ve been experiencing. We actually received an email from the Virtuozzo CEO earlier today. The email itself is long, but I want to show you one part of it:

“Finally, we continue to work to address the impact of the Spectre and Meltdown hardware-based vulnerabilities. Our priority is to ensure our customers have the latest information on the fixes and potential impact to your business. We know one of the biggest concerns has been around the performance issues associated with these fixes. We have been addressing the vulnerabilities proactively by sending email updates to subscribers. We recently released kernel with retpoline for Virtuozzo 6 which improves performance to near pre-Spectre level. A similar update for Virtuozzo 7 will soon be available. Look for additional email communication soon that will detail the results of this update.”

(We are using Virtuozzo 7).

You mentioned experiencing these issues for the past few months, it would coincide with the Spectre/Meltdown mitigations that were released in January/February. We are working hard to get these resolved.


Matt Connor
Founder & CEO
Strasmore, Inc.

I am still dealing with their tech support. They want to be able to reproduce the problems that I am seeing. I’m not sure what the best way to do this would be.

Hi @Linhead. Thats pretty much what I’m seeing on my VPS although not stuttering as much as freezing for extended periods. My local server (running Rclone cache works without fault).

@mjmayer can I suggest that you open a support ticket with SSDnodes and let them know the issues you are having. It might also be helpful to to reference my ticket number 396020

I think that it will add weight to the issues if a number of us are reporting these problems

@Animosity022 I’m not sure how to test for bandwidth issues either.

I just ran a pretty informal test for download speed.

1 Using Midnight Commander GUI file manager. A simple copy from my decrypted Rclone Cache to /tmp. Watching the transfer in iftop I see 5 or 6 connections to GoogleDrive and it very quickly gets to a consistent 330 Mbps and stays there until the transfer (a 2GB file) is complete.

2 I copy a different file from the same directory to /tmp but this time I use the following command.

sudo rclone copy local-crypt:/SSDnodes/ssdnodes-server1.2017-10-29.tar.bz2 /tmp

Again watching in iftop I see one (possibly two but only briefly) active connections to GoogleDrive. The speed fluctuates wildly going up to 200Mbps but then dropping as low as 20 (mostly below 60). This continues until the transfer is complete. I didn’t time the transfer but it took a lot longer for a similarly sized (2GB) file.

Shouldn’t the two copy functions work the same? Is this a bug in Rclone?

Sorry as I wasn’t quite clear.

I’m good on how to test speeds of a file copy but if the provider is throttling your VPS, how would you know that without them telling you? That would be seen as inconsistent speeds and odd things happening which it looks multiple users of that service are reporting.

I have a bit more control over my connection so I can validate via ntopng connections and what else is consuming my connection. I can run the config I shared streaming multiple 4K streams without stuttering or freezes in the frames.

Right but don’t you think its odd that using rclone copy i’m only getting one download stream from GoogleDrive and then with copy in Midnight Commander I’m getting four or five times that?

I’ve all but given up on my SSD Nodes VPS. They are overselling their servers & performance over time deteriorates then when you complain they do a fix which works for a bit then the performance starts to decline all over again. I have never seen the network bandwidth remotely approach the promised 1Gbps. I paid up front for 12 months of an X-Large VPS but will not be renewing asI think the performance is crap & their overselling deceitful.

Who are you planning to use instead?