(no subject)
Feb. 19th, 2005 12:51 amWhy do cool shows have to break servers? ;_;
I ended up having to shut down Apache for a bit tonight to be able to get into the server. Folks pulling LAY down were just puting too much load on the server. Its not really a money issue, its a hardware/bandwidth issue. The server's been locked at a constant 800kbit/sec for the last two days o_O.
Here's the weekly graph, taken as of 1AM EST Saturday morning:

I'm really glad people like the show so much, I just can't handle that kinda load.. Sorry! ^^; I think I've seen it suggested of only posting one show a week, and I think the server should be able to handle that.. Hopefully someday I'll get the server up to spec where it'd be able to handle this load.. I'd like to post more videos!
(Just for note, I'm not mad at all for people posting this show. There's no way they could have known how popular it would be ^^)
I ended up having to shut down Apache for a bit tonight to be able to get into the server. Folks pulling LAY down were just puting too much load on the server. Its not really a money issue, its a hardware/bandwidth issue. The server's been locked at a constant 800kbit/sec for the last two days o_O.
Here's the weekly graph, taken as of 1AM EST Saturday morning:

I'm really glad people like the show so much, I just can't handle that kinda load.. Sorry! ^^; I think I've seen it suggested of only posting one show a week, and I think the server should be able to handle that.. Hopefully someday I'll get the server up to spec where it'd be able to handle this load.. I'd like to post more videos!
(Just for note, I'm not mad at all for people posting this show. There's no way they could have known how popular it would be ^^)
no subject
Date: 2005-02-19 06:10 am (UTC)I've been using tc on my Linux firewall for awhile now and it works well - I can have my net connection pegged, and yet experience essentially no delay on ssh sessions.
no subject
Date: 2005-02-19 06:11 am (UTC)I'm running Apache 1.3 and Debian STABLE
no subject
Date: 2005-02-19 06:36 am (UTC)For the Linux traffic control, I'm doing it on a 2.6 kernel, but unless I'm mistaken, the code to do it has been there since 2.2 kernels! You'd need the 'tc' command from the iproute (iproute2) package to control it, and it's used for setting up traffic control in the kernel.
In my case, I'm making use of all four aspects of tc - shaping (control the rate of outbound traffic), scheduling (prioritize outbound packets), policing (similar to shaping, but for inbound packets) and dropping (if packets arrive too fast, I drop some of them to force the window size to be dropped, thus slowing traffic down, and same for outbound packets that exceed the buffer)
In my case, I have ... 4 outbound queues setup:
Priority (ie: interactive traffic, small packets)
Normal (default)
Reduced (ie: bittorrent downloads)
Minimum (ie: bittorrent uploads)
The queues are allowed to use 100%, 90%, 50% and 50% of my available bandwidth, respectively, and no packets are sent from a queue unless all higher priority queues are empty. The queues themselves use the Stochastic Fairness queue algorithm which tries to let all connections get bandwidth by sending packets in a round-robin fashion based on flow (flow being defined here by source IP, dest IP, source port) - so it'll send one packet per unique triplet as long as there are more than one set of packets waiting.
It's working pretty well, although I need to tweak it. It's heavily based on the 'wondershaper' setup right now, but tweaked to my environment.
SFQ (by default) only can handle a queue of 128 packets total, so if more than that many packets back up, it drops the most recent ones from the connection (as defined by the above triplet) with the most packets queued. Generally, this means I don't drop packets except when using bittorrent, and then it's only on the lowest priority, upload queue.
In woody, you want the iproute package, plus the ipmenu and/or shaper packages might be useful (though I have no experience with either, myself)
And, like I said, tc is cryptic, and not well documented when it comes to the rules for how to matching packets.
no subject
Date: 2005-02-19 06:27 am (UTC)no subject
Date: 2005-02-19 11:36 am (UTC)And incidently Shadow, that graph looks like a standard mrtg graph, but I'll let Inaki confirm that.
no subject
Date: 2005-02-19 11:39 am (UTC)