Shared buffers, remote buffers or pproxy?

[Slapstick]

Quick question, In the last 2 days I have added 3 PII 450's to the herd in order to work a few bugs out before adding more. These are running on a small Novell 5 network (It's great being the admin, I get to break my own rules about adding software), anyway, at the moment I have them using remote buffers (only one client has full internet access all the time). Would it be better have them share buffers, keep using remote buffers or to setup a pproxy? I'm leaning more towards keeping remote buffering or setting up a pproxy since I can set the client to buffer enough work locally to keep them busy in the event of a network failure.

Hope have a total of 7 to10 PII 450's working at least part time in the next few days with the possibility of adding several more at a later date.

 

[Engineer]

Slapstick,

I set up Remote Buffers at home in sort of a round robin fashion, although this may be too much for your needs. I have 3 computers (all have net connections, but If having only one, will still work).. I pointed the remote buffers of #1 PC to #2 PC, #2 PC to #3 PC and #3 PC to #1 PC...I then set each to load 2000 WU's and set the keyserver to the team PProxy. If the net goes does, each has a nice block of 2000 WU to chew on, and if one runs out, they will rotate the block back and forth to each other until they run out completely....

I have the Fetch/Flush threshold set to every 30 minutes to make sure they don't run low as long as there is an internet connection present...

Good Luck!

 

[RaySun2Be]

Slapstick,
I used to use Remote Buffers, but started to have issues with file locking as well as the newer clients fetch/flush methodologies. I went to a proxy and I'm glad I did. I definitely recommend using a proxy.

There are stats utils that you can use to look at your herd's productivity that way, and then you can go slap some user for running apps that are taking up CPU time away from RC5.

Ray

 

[Slapstick]

Thanks for the replies

Engineer,

I have 3 at home and just added a router so all can have net access but I like the way you have yours set up at home to keep them all running if the net connection goes down. I'll have to set it up that way when I get home tonight.

RaySun2Be

Sounds like a pproxy is the way to go at work, I wasn't aware that there might be problems with the new clients fetching/flushing.blocks using remote buffers. Got any tips on setting up the pproxy? Need to ensure enough blocks so the cows wont go hungry over the weekends.

On a side note had to update printer drivers on 2 PC's over lunch so make that 5 added today, (now if I can only keep them from turning off their PC's at night). Of course since I set up a generic user on the network who is restricted to accessing just the directory the buffers are stored in I could just log their PC's back in

 

[Kilowatt]

Slapstick, I have my pproxy at work, (feeding 5 dualies, 6 workstations and my 4 node rack) set for 20,000 work units.
It's set to update to dnet every 2500 WUs done, as to stay full.
In the event the net connection goes down, it'll last about 2 days before running empty.

Even if your net connection goes down, your LAN should still be working, and your herd should stay in feed untill it's fixed.
If the net connection doesn't getf fixed within 2 days, I'll have to sneeky in a load from home.

 

[Viztech]

All excellent advice above.
I might add to Engineer's scenario that the PC with net connection will attempt to connect to the net first before it falls back to remote buffers, making his solution most ingenious

I have had trouble with buffer locks using remaote buffers with a lot of cows (10-15) trying to share when cracking RC5. Problem hasn't happened with OGR due to the much lower traffic.

Also have had the trouble you mentioned with with directory permissions using remote buffers. This is not an issue with pproxy, but you must be running TCP/IP to use a pproxy.

PProxy would be the best for you, remote buffers are OK if you can tolerate buffer lock-ups on occasion.

Good luck!

viz

 

[Engineer]

Thanks Viztech! A great complement coming from you...

P.S. All PC's at home have the net connection through the router also, and the remote buffering just acts like a good backup. When going on vacation, I'll stockpile one of the PC's with 10-20K blocks in case the router locks up or the net goes down...

And I too have had a few remote buffer glitches at work (many PC's), but never any trouble at home at all...

 

[RC]

Now that I am increasing the number of machines at home I also will need to setup a pproxy on my home LAN. I found this thread very informative. Thanks.

 

[Viztech]

RC-

Yes, these threads can get really informative sometimes, especially on these very issues.

Engineer-

Your welcome, but if the shoe fits...
I remember having a thread a while back where we discussed this very subject. I did not try it until the other day when one of my site's ISP was down. One cow was chewing on the last packet, while the one next to it had 2 days left. Since these machines all share the Inet connection and the clients normally flush direct to Jator, I was not using remote buffers. I added remote buffer updating to the machine that was hungry, and created a share on the one with extra feed. When I restarted the client, it tried to connect with Jator. After 3 attempts, it updated via the remote buffers and all was well. It was sure neat to watch this great idea in action during a crisis.

viz

 

[RC]

This combination of pproxy and remote buffers would not work on a diskless crackrack, correct? Use of a pproxy is the only alternative, right?

 

[Engineer]

RC, unless you used a floppy for your buffers (LOL!)....The pproxy would probably be the best bet for such a beast..(Well, I hope it will be a beast)...

Any others on this idea?

 

[SpaceWalker]

pProxy is definately the way to go....

I'm running 2 proxies at the moment, one at home(3 computers) and one at work(2 computers right now), Both flushing thru Jator's proxy....


MOOOOOOOOOOOOOOOoooooooooooooooooo..............

 

[KingHam]

I also use proxies. One at home and one at work. Management has become tons easier since I switched to proxies.

KingHam

 

[RC]

Let's just say it will be a 'noticeable' improvement over my current keyrate -

 

[RC]

For a node that is not diskless I'm not sure I see all that much to be gained by using a pproxy. One can always set the buffer level to a very large value. This way if the machine loses its connection to internet it will simply keep fetching from its large buff-in. There does exist the freak incident where perhaps the connect to internet may be down for a week. If I fail to connect to my ISP for a week I will be changing ISPs -

One can use one of the existing pproxies (such as Jators) to be able to monitor the health of the individual cows.

If you have multiple cows on the same IP then a pproxy within the LAN could be beneficial.

These are just my ramblings -

 

[Kilowatt]

RC, You should run the Rack's pproxy on a "Server" machine running either WinBlow$ or Linux, makes no differance.
The pproxy can server any OS version of the clients.

That way, all machines flush/fetch their blocks from the same "Pool".
Having the pproxy on your everyday machine that connects to the internet regularly will keep the in-buffer filled and the out-buffer flushed.

There are some pproxy monitoring programs for Unix, Win32 and NT, that will let you see what each Node or client is doing.
Some are better than others, some are lighter than others.
Some you can just run when you want to take a look at what their doing and shut off again.

The one I use is for NT, and I've got it set to update (ftp) it's stats.html page to my webspace every 15 min.

 

[RC]

kilowatt - yep, I will have to use a pproxy for my crackrack, that is a given. Besides, where would the client on the diskless node save blocks in the buff-out.

Question for you on klinux and the dnet client: How do I configure the client so that all completed WUs are sent immediately to the pproxy. If the diskless node loses its network connection what does it do with completed WUs?

[edit] I believe the diskless node would keep the completed WUs in memory until a network connection is available. [/edit]