Linode availability and i/o token questions
I've been eyeing up Linode as a place to host a VPS, specifically a Drupal installation of mine that will require SSL. I've looked through the features, the screenshots, and the forums here, and I am happy with what I see.
I've a few questions though:
1) What's up with availability? It seems that any linode above a 360 has been unavailable for the last few weeks. A few of them were estimated to be available today, March 23rd, but now the date has been pushed back to March 30th. This brings me to my next question…
2) Are there any hot spares available? My concern is that a hardware failure might cause one or more Linodes to become unavailable for an extended period of time while new hardware is ordered. I'm not implying that Linode would ever let this happen, but I need to ask, as this scenario happened at a past virtual machine provider that I did business with, which resulted in days of downtime for those affected.
3) The i/o tokens. I understand how they work, but I'm a bit confused as to how fast they get depleted. Can any folks out there tell me how many tokens might be taken up by say, a compile of Apache or PHP? Or perhaps dding a 100 MB file with input from /dev/null? Basically, I'm trying to keep something like this
Thanks for your time,
– Doug
3 Replies
caker
Linode Staff
We've been working hard trying to keep up with demand… We have 10 servers shipping out this week, all of them should be online by mid to end of next week. (I'm actually in the office right now checking in on the burn-in)…
We keep at least one hot spare machine in each datacenter at all times. They are there for a reason.
The token-limiter isn't there to prevent you from performing normal work -- a Linode should only hit it if it's doing something bad like thrashing swap. Swap thrashing has become less and less of an issue as we've increased the resources on the plans, and also since our disk subsystem is a lot faster than when we started out. Both of those have allowed us to increase the default settings for the token limiter much higher than they previously were. So, it gets triggered less often. It's really only there as a brake for run-away Linodes.
To answer your question, one token represents one IO operation issued from your Linode. You get a refill of 512 tokens into your "leaky bucket" per second. For reference, most Linodes average in the 0-10 I/O ops/sec range.
-Chris
Thanks for the quick response, and I'll keep an eye on the availability.
– Doug
@caker:
We keep at least one hot spare machine in each datacenter at all times. They are there for a reason.
…. Swap thrashing has become less and less of an issue as we've increased the resources on the plans, and also since our disk subsystem is a lot faster than when we started out. Both of those have allowed us to increase the default settings for the token limiter much higher than they previously were.
….
-Chris
Depending on the model/scale you guys are going for, you might want to look at cheap central storage running SOFS or other similar things (EMC has an offering, not sure about others off the top of my head) to either house the 'data' or possibly the VM's themselves. I've been meaning to type up a blog post about this (scalable distributed 'intelligent' filesystems) for a few weeks …. just a little too busy.