This post will just apply to users of the “bears” server. I’m having a personal issue with the way DreamHost has redefined the notion of “overloaded”, and wondered if anyone has encountered the same kind of response.
“bears”, according to one of the tech support representatives, is a 16-core CPU server. The last time I’ve checked, it had around a thousand users and at least that many websites. Until around early February, average load of that server would be between 4 (on the low traffic hours, i.e. early Californian morning) and perhaps 6 or 7. This is about 40% workload, and it meant rather responsive sites.
With the load above 12, it means “a few seconds”. When it rises to 16 (100% CPU load for 16 cores), it means a lot of seconds. Really a lot. A minute or so to open a post/page to update it is not unheard of.
So I reported the unusually high load, and a tech said something like “oops, something is not right; but someone is watching the high load and sooner or later will do something about it”. This is usually attributed to some rogue processes, or user configuration mistakes (often my own fault! But I get instructions to fix that). In some cases, it might be a DDoS attack, and in that case, we’ll get notice of it on www.dreamhoststatus.com. In other cases, there might be some problems with the hardware itself or the way the disks are mounted, etc. Or it might be something unusually hard to track down and fix.
In any case, 2 or 3 days after my report, something got fixed. The load dropped to the usual 4-7 as before. It remained like that for a few days… and then climbed up very quickly again. It reached 16 in no time, spiking to 19-20; the lowest I caught was around 12 on the early morning hours (which would mean 75% workload). Since then, I’ve seen it spiking up to 26 (!) and above. That’s not good at all, and much worse than what I had reported a few weeks ago.
So I reported it again.
This time, however, I got a completely different tone in the reply. The support techie assigned to my case looked that the load was 12, and since that was below the 100% workload of 16, he said “it’s normal for a 16-core server”. Really?
Well, perhaps. Apparently, these days system administrators tailor their systems for running at 70-75% workload. I seriously suspect this trend has started with Apple’s Mac OS X, which is almost always close to 60-70% workload, and it’s considered “normal”. Even so, I think I could “live” with a load of 12 on average, so long as most spikes would be below 16 or so. The problem with “bears” these days is that it averages on 16 or so, but it runs over 20 most of the day (just right now, it has peaked at 30 (!)).
So I wonder about two things:
If you are hosted on “bears”, how has your performance been? Performance can be misleading because we have different expectations. My expectation of getting every WP backoffice page in a second or so might be unrealistic, but that’s what I tailor my own servers for — anything more than that, and my users will complain. DH’s users might be more tolerant and be fine with waiting a minute for their backoffice pages to load…
If you have complained to DreamHost’s tech support, what answer did you get? Did they politely ignored your request saying that it’s “normal” for “bears” to have workloads of 20 or even 30?
I’m just curious. For the first time in many, many years, my sites on GoDaddy, which has a legendary bad performance on their own servers, are actually faster and more responsive than DreamHost. I’m utterly shocked at that, to be honest. Sure, DH is slightly less performing than many of their competitors, but it’s a shame to be compared to the hosting provider at the bottom of the list and be even worse than that. Also, it makes it rather hard for me to explain to some customers that the pitiful performance they have been experiencing in the past few weeks are considered by my hosting provider as “absolutely normal”, when for years the performance has been more than adequate. They question my skills in picking a good hosting provider!