I'm curious how people are getting along with the Linode 96 and 128 products. My experience with the Linode 64 is less than satisfactory. I will attest that you can't run anything reliably on them except for maybe a low traffic DNS server.

I'm considering upgrading, but I'm not convinced this UML stuff is much better than an old 486-66 sitting on a rack.

Thanks
-infinity

My Linode 64 is happily running about 7 websites. One of the websites is moving several Gb per month with plenty of PHP and MySQL activity. It also runs sendmail, pop and Spamassassin for about 5 people receiving lots of mail.

Load averages seldom show above 0.00 when I look. It has a 128 Mb swap partition which usually has about 60 Mb free so everything is running fairly calmly.

I think your experience on host12 is not the normal case.

Here are the list of current issues:

Host11 and Host12 are not running the latest 2.4.23-1 kernel that I deployed to the rest of the hosts this past week. The new kernel has some interactivity and I/O balancing patches which so far have shown a good improvement. It's time to upgrade host11 and host12.

High levels of interrupts caused by the network device -- I'm going to assume this has contributed to some network lag or other weirdness that some people have reported. I'm currently testing enabling NAPI Rx Polling, which greatly reduces ints generated by the NIC.

There exists some weirdness or perhaps bug inside UML (or in Linux, for that matter), or something that someone is doing with their Linode that can cause the host to slow down. I've seen one Linode fill up the disk I/O queue on the host, but without many blocks in or out from the raid. Sounds like it's pounding the host's page cache, but the numbers aren't very descriptive. This is something that I monitor for and take care of usually in a very short amount of time.

I'm working on a patch to UML that limits the number of requests under high load.

Also, Jeff Dike is working on two features which will give HUGE performance boost to memory and I/O operations inside UML. /dev/anon and ubd mmap. "ubd mmap" maps the ubd device into the host's page cache directly, and /dev/anon is a new memory driver for UML that keeps UML from keeping copies of data in its page cache, as well. What does this mean? Less copies of data and no more wasting host memory when it could be going to a larger cache.

And if that's not enough, when Linux 2.6 is stable enough to run on the host, we can use the Completely Fair Queuing (CFQ) disk scheduler -- which round-robins the disk queue.

I'm serious about these performance problems that crop up every now and then. This is my number one priority.

-Chris

I have been on a Linode 64 host (host5) for about 3 months now and have had very few problems. Slowdows and network outages don't seem to be affecting this host. I don't know why some hosts have different configurations than others, but whatever host5 has, it seems to work just fine.

This explains the slowdowns I've been seeing on my Linode on host12. It's fine most of the time, but every now and then it slows down horribly.

What's interesting is that the load average as seen from in my Linode shoots way up even though the system is very lightly loaded. Once, the load average went up above 8 even though all I was running on there was top.

Are you planning to install the updated kernel on host11 and host12?

My linode 64 on host5 has run just fine, too.

DNS server and mail hub for several domains, LNX-BBC garchive (updated thrice daily), and LNX-BBC nightly build archive for CVS HEAD.

Yes, I'll be sending out the announcement today or tomorrow -- scheduling the reboot for Monday evening, most likely.

I also discovered we're receiving traffic from HE that doesn't belong to us. Averaging 2-12 Mbits/sec which is what I think is the cause of the high interrupt rate (lots of small packets). I'm awaiting HE to resolve this issue on their end, hopefully today. If not, I'll activate some ACLs on our switch to filter the traffic to the host machines.

-Chris

Can I get moved off host12 then? I doubt I'm causing the problem with the box, but who knows!

You can move me anytime. I'll be happy to get off that 486. ;)

You asked for it -- I moved you to host14. You're the only soul on the machine at the moment.

I'm curious, why do you call it a 486? For the past few days the load has looked great on all the hosts, including host12. This post also outlines a correction HE made over the weekend.

What type of workload are you doing on your Linode?

-Chris