Missing blocks from this week? Here may be why...

[Choralone]

I just received this .plan update in my e-mail. (I'm on D.net's e-mail list) Check the second paragraph.

bovine :: 03-Mar-2001 09:05 (Saturday) ::

After many delays we are now proud to announce the availability of the
official line of distributed.net-branded clothing and merchandise,
affectionately known as "dnet-ware". We are currently offering two styles
of t-shirts, one style of sweatshirt, coffee mugs, and mouse pads. Feel
free to cheeck out the available items on our "dnet-ware" web page at
http://www.distributed.net/dnetware/

Additionally, over the last few days we have been undergoing a fair amount
of workunit submission growth, which has been causing a backlog of workunits
to accumulate at our keymaster. In order to improve responsiveness, much
of this backlog has been transferred offline for gradual processing over
the next few days. Some new optimizations to the keymaster build 324
codebase have also been made to greatly improve sustainable workunit
processing rates by several fold under most circumstances. We will continue
to be processing the backlog over this weekend and stats should hopefully
be caught up and up to date within the next couple of days.

So it looks like some of the work that had been submitted earlier this week still hasn't been counted but will be soon.

 

[Poof]

Thanks for the update Choralone!

I would think that maybe upgrading the K62/350 master keyserver WOULD BE a better idea at some point... Sigh... :frown:

 

[Jator]

The keyserver is a K6-2 350??? You are kidding, right?!? Man, if that's the case, I'm sure we can donate enough money to purchase a PIII 650 for them.

 

[Viztech]

Hmmm, from info found on Dnet's history page, it's a K6-2 300.


<< July 11, 1998 The new keymaster goes online, sporting a K6-300, 128meg SDRAM, 4.5GB UWSCSI, 1.2 GB USCSI, 6.4GB IDE, and running FreeBSD 2.2.6-STABLE. The machine booted and began serving keys without incident and ran without reboot or failure for nearly 200 days. >>


viz

 

[Poof]

jator &amp; viz - I got the &quot;350&quot; figure from one of Moose's posts in a client recall thread... Looks like it's worse than even he thought ...

It's a wonder it's still functioning!

 

[Choralone]

If I remember correctly their K6-2 350 is able to process quite a bit more than even D.net's heaviest days of dumping. Provided all the blocks that are submitted are in the 2 blocks of keyspace they keep open on the master. I thought a saw a figure somewhere in the range of excess of 100+ million blocks a day.

Each block of keyspace is loaded into RAM and is approx. 32MB. So two blocks of keyspace would consume 64MB of RAM and the OS overhead and other programs I'm sure eat up most of the remaining 64MB. What slows the master down is when it is checking in blocks that are NOT in the 2 open blocks of keyspace. Those blocks are ones people have been hanging onto for a while and randoms.

I'm not sure exactly how D.net has things setup but either the master unloads one of the 2 open blocks, loads whatever block the key belongs to, and then unloads the block that it just loaded and re-loads the block of keyspace that should be open. Or it loads 3 or more keyspaces up and uses the HDD as virtual memory (does FreeBSD do that?).

Either way you're talking about MASSIVE disk I/O. :frown: No matter how fast their HDD's are (I noticed they're pretty fast already) a HDD will always be slower than your system RAM.

If the above situations are true then a memory upgrade would be more beneficial to their master keyserver. Although a faster CPU wouldn't hurt either. I know from experience what moving my machine from 128MB to 512MB can do!

 

[Adul]

Why stop at 512 MB of ram? I say with 1.5 GB worth. That should be more then enough for a stats server.

 

[BurntKooshie]

They have way more than enough funding....they have over 20k just sitting there. We could donate the cash to make it, but I doubt they'd buy more stuff to beef it up.

 

[Choralone]

Adul - It's not the stats server that's having trouble keeping up (although it has enough problems of it's own) it's the master keyserver. But, the master keyserver has to be able to do it's thing before the stats server can feed us stats. Plus with a K6-2 based board I don't think you could put anything more than 512MB of RAM in it. I know some of those boards won't see a 256MB DIMM.

BK - I don't think it's a money issue keeping D.net from upgrading the master keyserver. I would imagine it's two things 1. The machine they have CAN keep up %99 of the time, it's the %1 that it doesn't when we notice. 2. It's probably the machine they want to take down the least out of all. Who knows exactly how things would go if they got a bad stick of memory or if they built a new machine and it developed a &quot;quirk&quot; after running for a month or two.

It is nice to know they're working on tweaking the software they run on it to allow the machine to process more work than it can now.

 

[networkman]

RE: The K6-2/350 machine

I've already got an email into Moose about this issue, and my email was followed up by another from mechbgon(as I'd cc'd him on my email); hopefully, Moose will have an answer for us shortly on our offer of equipment and/or funds to do a major upgrade to that box. I know that similar offers of donations have been made by our DPC counterparts making this much more of a combined effort than a single team issue.

Let us hope that they will accept our offer of help, so that MEGA-flushes and increased growth will not be frowned on!

 

[BurntKooshie]

choralane - that was my point There's been quite a bit of support of giving them additional help, whether that be via more hardware, or cash for hardware, but the fact of the matter is that's not their problem.