4:45 AM UTC, 9/7/14

Rather than start a separate thread, I will note here that GPU72's Assignments page is responding with:
[QUOTE][B]500 Internal Server Error[/B]

[B][SIZE=2]Internal Server Error[/SIZE][/B]

The server encountered an internal error or misconfiguration and was unable to complete your request.
Please contact the server administrator, [EMAIL="chalsall@ideas4lease.com"]chalsall@ideas4lease.com[/EMAIL] and inform them of the time the error occurred, and anything you might have done that may have caused the error.
More information about this error may be available in the server error log.
[/QUOTE]I'll also drop Chris an email.

[QUOTE=kladner;382354]4:45 AM UTC, 9/7/14

Rather than start a separate thread, I will note here that GPU72's Assignments page is responding with [...]
I'll also drop Chris an email.[/QUOTE]

Since the manual submission page got a reworking, and that's what GPU72 uses, I'm going to guess Chris might need to tweak something in there.

I haven't looked at the new manual pages myself so I don't have any idea what changes might be needed, if any.

The page is back to normal now. 5:40 UTC

Any feedback on the performance?
 

Besides changes on the software side of things, how are y'all liking the server performance itself?

Just to recap, we've had 2 different migrations.

First one was from the old hardware to a temporary virtual server. That was on Aug 19/20 and involved most of the heavy lifting with upgrading the OS, SQL, PHP, and fixing up things that needed it as a result of the updates.

That temp virtual machine had 16 GB of RAM and had a C: and D: drive for the SQL data/log, and it did pretty well for itself.

The second migration is the one we did Thursday (Sep 4) to it's new server. It's still a virtual machine for easier migrations in the future, but it's on a dedicated physical box.

The new setup has 54 GB (I saved 6 for the host OS) and has dedicated virtual drives, each on their own physical disks, for the SQL data, log, and now a separate array just for the indexes.

With that amount of RAM, the entire DB could basically fit in memory, and over time SQL would do just that. Disk reads and writes are still fast, but having things cached in RAM is a pretty big deal. :)

I wanted to share some of those nitty gritty details for the geeks out there, and also to get any feedback on the "feel" of the site. It's one thing to look at a bunch of numbers and graphs, but the "perceived" performance is what matters.

All of the GIMPS participants thank you and also of course Scott, James, George and ... !

I do not use the site a lot these days, but it does seem much more responsive.

Thanks again.

Jacob

Madpoo, you could talk a bit about power (consumption, monitoring, and managing). I know nothing about it.

About the data base, physical disks, tables, indices, and logs .... I cut my teeth on data bases, but that was back in 1980. At the time, I got best performance by logging to a dedicated device (because logs are written sequentially)

Index files went to a device other than their table. In other words, not all indices to one device and all tables to a different device. That was just a rule of thumb, an initial allocation before profiling, balancing, and optimizing. It was also in the day when a 4 MB drive cost a fortune and memory was at a premium.

[QUOTE=M29;382431]Madpoo, you could talk a bit about power (consumption, monitoring, and managing). I know nothing about it.

About the data base, physical disks, tables, indices, and logs .... I cut my teeth on data bases, but that was back in 1980. At the time, I got best performance by logging to a dedicated device (because logs are written sequentially)

Index files went to a device other than their table. In other words, not all indices to one device and all tables to a different device. That was just a rule of thumb, an initial allocation before profiling, balancing, and optimizing. It was also in the day when a 4 MB drive cost a fortune and memory was at a premium.[/QUOTE]

The colocation the server is in is providing 312W (1.5A @ 208V), so I had to enable the power capping feature on the Proliant server at 312W. In truth, capping it didn't really do much. To test it, I used Prime95 running on all cores and then checked how much power it drew if left unrestricted, and it went up past 400W.

Capped at 312W, all it really had to do was set the processor cores into P1 state, basically turning off the turbo boost. If I set the cap even lower like 250W it would throttle the cores even more, down to P3/P4 territory, lowering the 2.67GHz CPU's down to below 2GHz, but that was enough to keep things under the limit.

In practice, the server doesn't use that much CPU and it's power use is under the cap anyway.

On the DB side, if you consider that the old physical server had everything crammed onto a single pair of 72GB drives (OS, web files, SQL data/log/index), it's easy to appreciate the new setup. The OS and web files are still sharing a drive with the SQL data (MDF) files, but since those tend to be more read-heavy, and it's a 4-disk RAID 5, it has no problem keeping up.

The separate 2-drive RAID 1 arrays for the SQL logs and indexes are also easily able to keep up with the write demands. Moving the indexes from one table in particular, which is basically a log of each client transaction, was a big help. The index is bigger than the table itself by a good amount, and it does get updated all through a typical day, so it kind of blew the whole concept of the data file being mostly reads... the index on it made it more write-heavy.

It also helps out when the index is being reorganized. Actually I haven't done that yet since moving to the new system because they were all rebuilt at the time, but in theory, it should make the reorg run a lot faster.

How much interaction have you have with Scott with this makeover?

[QUOTE=M29;382431]Madpoo, you could talk a bit about power (consumption, monitoring, and managing). I know nothing about it.[/QUOTE]

On the monitoring side, for now I've added the host and guest OS to my own monitoring suite (Solarwinds NPM). It's pulling system stats every 15 minutes. I'll probably set some alerts to ping me when drive usage gets high or if the CPU starts spiking for prolonged periods.

Managing it for now will probably just be standard stuff... make sure it's patched with the latest Microsoft and PHP things, ensuring backups are happening, and generally following the "proper care and feeding of a Windows server" instruction book. :smile:

Let's see if I can successfully attach a PNG that shows the ambient server temp for the past 24 hours. Long term I probably can't keep monitoring the system with my own software so we'll find something that can do the basic monitoring and alerting. Disk space is probably tops on the list since there have been several examples of the log file filling the drive and causing errors. But the Solarwinds product is [B]so[/B] cool, I had to make use of it while I can. :)

[QUOTE=Xyzzy;382449]How much interaction have you have with Scott with this makeover?[/QUOTE]

He and the rest of the gang (George, Luke, et al) were involved in a lot of the decision making like the selection process for a new colocation for the server. As far as hands on it was basically George, James and myself.

I hope they had as much fun as I did. Upgrading a system like this and being able to track some real time progress is kind of fun.

For instance, we updated the robots.txt file the other day and there was an almost immediate improvement in the # of instances where [url]www.mersenne.org[/url] is showing up in someone's search results. That plus the performance increases gives me that warm, satisfied feeling of a job well done. :smile:

[QUOTE=Madpoo;382471]He and the rest of the gang (George, Luke, et al) were involved in a lot of the decision making like the selection process for a new colocation for the server. As far as hands on it was basically George, James and myself.

[B]I hope they had as much fun as I did. Upgrading a system like this and being able to track some real time progress is kind of fun.*[/B]

For instance, we updated the robots.txt file the other day and there was an almost immediate improvement in the # of instances where [URL="http://www.mersenne.org"]www.mersenne.org[/URL] is showing up in someone's search results. That plus the performance increases gives me that warm, satisfied feeling of a job well done. :smile:[/QUOTE]

*It has been fascinating and educational to follow the discussions, and to watch the whole enterprise come into focus. Many thanks to all who put thought and effort into this accomplishment.