New servers: what is reasonable load average?


My account was moved to its new server yesterday. Load average is in the range 5 to 10. Maybe this will increase as more accounts get moved on to it. How many cores do the new servers have, and what is a reasonable load average to expect?

It appears that Dreamhost didn’t bother with reducing the TTL before doing the move. I’m not saying they absolutely should have done, this is more of a question really … should they have?

Operating system on new server is Debian GNU/Linux 6.0.9 squeeze (on old server it was version 6.0.5)

P.S. no apache glitch slowdown observed so far.


The new server I’m on has 8 cores. get in ‘top’ and hit 1 to check yours. Your load sounds about avg.

Dreamhost leaves the old server up and serving at the old IP plenty long after the move. The sites accessible from old or new until dns has had plenty of time to catch up, that’s why they don’t bother lowering the TTL.


Well that’s exactly what I’m wondering about.

There were a couple of hours during which my website was live on both the old and the new server, and a browser would hit whichever one its DNS routing happened to take it to at that moment.

One my home machine I arrived at one version, and using TOR (the onion router) I arrived at the other version. (I verified this by changing one of the versions.)

What I’m wondering is, shouldn’t this window of potential inconsistency be made as short as possible, via TTL management?

Because during this time, two separate installations of the php files are accessing the same tables in the same database. I don’t know if that’s ok, it just feels like the sort of thing that could cause problems.


A week on in. Slight but observable ramp up in the load average on my new server.

For the first few days it was in the range 5 to 10 and now I’d describe it as being in the range 8 to 15 (with a spike of 145).

I guess it is a good thing if a cross section of customers report their load averages from time to time.

No sign of the dreaded apache glitch so far on new server.

As an entertaining tidbit, ‘wc -l /etc/passwd’ says 1277 as an estimate of the number of users on the server. This may be fairly meaningless as a measure of workload since many or most users could be mainly quiescent but it is worth noting.


As an update on this, I have a cron job doing ‘uptime’ periodically on my new server. Here are the average values of the last-15-minutes ‘load average’ figure for each month since May:

2014 June 8.93
2014 July 8.35
2014 Aug 7.70
2014 Sep 7.21
2014 Oct 9.18
2014 Nov 6.68
2014 Dec 4.40
2015 Jan (so far) 4.03

Drilling down into the results for November, a decisive change happened fairly clearly between 20:20 and 20:40 (times are in UTC) on 16th Nov:

19:00:02 up 28 days, 15:04, 0 users, load average: 11.96, 12.99, 13.05
19:20:02 up 28 days, 15:24, 0 users, load average: 11.00, 10.89, 11.05
19:40:02 up 28 days, 15:44, 0 users, load average: 25.06, 19.45, 15.65
20:00:02 up 28 days, 16:04, 0 users, load average: 18.49, 17.79, 17.51
20:20:01 up 28 days, 16:24, 0 users, load average: 12.67, 14.66, 15.67
20:40:01 up 28 days, 16:44, 0 users, load average: 2.80, 4.54, 9.13
21:00:02 up 28 days, 17:04, 0 users, load average: 4.04, 4.21, 5.55
21:20:02 up 28 days, 17:24, 0 users, load average: 3.02, 3.68, 4.51
21:40:01 up 28 days, 17:44, 0 users, load average: 4.30, 4.00, 4.08

Perhaps a misbehaving program was removed.

edited to say still no recurrence of the “dreaded apache glitch” which ran rampant at Dreamhost in 2013 and 2014.