Using just mining.eligius.st is depreciated, as the IP(s) it points to are not maintained actively anymore.
They should be, considering mining.eligius.st is just a CNAME for getwork.mining.eligius.st, which is a supported hostname...
In detail, all of the mining (and the web server) for Eligius is hosted on a set of servers on a private LAN. The private LAN is then tunneled out to multiple low-latency servers around the US. All of these are used for the stratum DNS record. The top few are used for the GBT record, and those with the least redundancy used for the getwork record.
Correct, mining.eligius.st was, up until a few hours ago, CNAME to getwork.mining.eligius.st. When the endpoints in the getwork record had some issues, even people mining with stratum on "mining.eligius.st" who did not update their hostname to "stratum.mining.eligius.st" also had issues. As of now, the getwork and gbt hosts alias the stratum hosts and have the same redundancy... for now.
In any case, eventually I want to get the getwork and existing version of GBT loads away from the stratum loads, hence the need for multiple host names. As evident by today's events, getwork and GBT (and stratum people on the wrong host name) account for about 20Th/sec of mining, or something like 3-4% of the pool. However, getwork and GBT account for a full third of the server load. From a system admin perspective, it would make sense to simply drop support entirely.
First, there is little to no reason that I can think of, and correct me if I'm wrong, for anyone to be using getwork anymore.
Before you start throwing exceptions at me, sure, some miners like the ASICMINER blades use getwork, but do not correctly support it. Pointing a blade directly at Eligius totally eliminates all pool side and client side efficiency gains from vardiff, since the blades do not check hash target and submit all difficulty 1 shares to the getwork server they are configured with regardless of what the server tells it to do. If configured directly to Eligius with out a proxy you're putting 4-8x more load on the pool than you should be per blade by forcing the pool to filter out your high hash shares and hardware errors. Use a local proxy, either bfgminer's or slush's stratum proxy and don't put this extra burden on the pool. Also, the blades are not multithreaded. Their mining software runs on a PIC microcontroller. Therefore, the blade waits for a response from the server before it continues and feeds new work to your hardware, thus lowering your hash rate proportionally with latency to the pool. Again, run a local proxy.
CPU mining is bitcoin is pretty much dead, except for some (probably now abandoned) botnets. When even 1000 CPU mining nodes makes a script kiddie something like $5/mo, even at current high prices, its time for a new hobby.
Getwork is good for dev testing... and I used it myself when testing out my mining patty prototype..... using a stratum proxy.
In any case, eventually the host names for various protocols *will* land on different physical servers. The new host names have been posted for quite a while now... please update to a proper one for your protocol of choice if you haven't already. If you don't have a protocol of choice, use stratum.
I want to continue to support all protocols as long as possible, and to do so they will eventually need to be partitioned better server side.
-wk