The cut & paste of the messages shown with debug on are: ---------------------------------- Timeout notAnswered=[0] notReady=[0] nrAns=0 nrRead=0 nrPart=1 avg=3574 nrBlocks=38 lastGoodBlock=0 --------------------------------- ................ ------------------------------------- main() 1 901 419 520 re-xmits=001812 ( 0.1%) slice=0032 73 709 551 615 - 0 main() 1 906 684 416 re-xmits=001812 ( 0.1%) slice=0032 73 709 551 615 - 0 bytes= 1 906 933 760 re-xmits=001812 ( 0.1%) slice=0032 73 709 551 615 - 0 -------------------------------------
I actually ran several installs using 10 clients with: * 50M Max_Bit_Rate (DL360-G3 systems) * FULL_DUPLEX = off * MIN_CLIENTS = 10 * MAX_WAIT = 800 * MIN_WAIT = 700 I found some peculiar behavior that may be one of you can explain to me. I power on all my clients (10) at the same time, and although most of the clients start their udp-receivers at close intervals, seems like the udp-sender establishes connection with most clients, but drops all of them but one (1st. Connected) and starts download to it. The remaining clients have to wait until the next window (after client 1) finishes its download. I thought that by setting the min_wait & max_wait to long values, it would fix the problem but apparently, it doesn't make much of a difference. Any ideas? Also, with only 10 clients, I am getting way too many TIMEOUTS: (i.e) Timeout notAnswered=[7] With tmpfs staging enabled, I would have thought that the clients would be able to handle a better data download rate. Why am I seeing timeouts with only 10 clients? I do not see timeouts at all, when I am installing 1 client at a time. I worry that when I start running SI over 256 clients at once, performance may go down. Carlos Ar�valo hp -----Original Message----- From: Brian Elliott Finley [mailto:[EMAIL PROTECTED] Sent: Friday, July 25, 2003 10:54 AM To: Arevalo, Carlos Cc: Frazier, Daniel Kent; SISuite Users Subject: Re: [Sisuite-devel] Multicasting Errors Hey, Carlos, I'm very glad to hear that it's working for you now. :-) Please see my comments below... Thus spake Arevalo, Carlos ([EMAIL PROTECTED]): > Hi Brian, > > I am happy to say that I am running multicast on several nodes now. > However, I see a lot of retransmits over the course of the image > upload, even when I am only installing 1 or 2 nodes. > Is this normal? Yes. > I am using 50M for my MAX_BIT_RATE Very good. > Also, while debug is enable I see a lot of messages right before the > retransmitts with some kind of TIMEOUT/INFORMATION messages separated > by TABS, (i.e AVERAGE, NR,) what do these different tabs mean? And > should I pay attention to them? Can you cut and paste some of these messages for me? > I want to have as much information as possible when I get to > manufacturing where I will be installing 256 nodes at once. > > One last question, with DEBUG disable, is there a way to check for the > multicast activity from the server side? Yes. Look in /var/log/systemimager/ for files called flamethrower* > Thanks again for all your help. Sure thing, Carlos. (more comments below) > Regards, > > Carlos Ar?valo > hp > HPTC > Phone: 603-884-3390 > Fax: 603-884-5191 > E-mail: [EMAIL PROTECTED] > > > > -----Original Message----- > From: Arevalo, Carlos > Sent: Tuesday, July 22, 2003 8:20 PM > To: 'Brian Elliott Finley' > Cc: Poirier, Keith > Subject: RE: [Sisuite-devel] Multicasting Errors > > > Hi Brian, > > Good news, it turns out that I should have added the tmpfs_size=<size> > on the APPEND line and not as a new line, Yes. Very good. > Plus I didn't restart the PXE server and that could have been the > reason for not resetting tmpfs to its bigger size. Restarting the pxe server should not be necessary. I want to build a small database of successful configurations. Can you send me your working flamethrower.conf file and a description of your hardware including switches and NICs? Also, after you've completed a large install (256+ nodes), please let me know how long it took. I'll add this to the FAQ at some point, and put your name next to it. Thanks! -Brian > So, now, I just finished installing 1 client via multicasting using > tmpfs staging of an image of size 1.5G. > > I will test installing 12 clients tomorrow to see if it works. > > I will hold on sending you the image and see if this works tomorrow > mornig. > > I really want to thank you for all your help, and I will keep you > posted of my progress (hopefully all good tomorrow! ) > > > Take care, > > Carlos Ar?valo > hp > HPTC > Phone: 603-884-3390 > Fax: 603-884-5191 > E-mail: [EMAIL PROTECTED] > > > > -----Original Message----- > From: Brian Elliott Finley [mailto:[EMAIL PROTECTED] > Sent: Tuesday, July 22, 2003 6:39 PM > To: Arevalo, Carlos > Cc: Brian Elliott Finley > Subject: Re: [Sisuite-devel] Multicasting Errors > > > The ramdisk is seperate from the tmpfs filesystem. It is used to get > the BOEL going, but the init script on the ram disk mounts tmpfs, then > switches to it as it's root. > > See my other mail for sizing options. > > -Brian > > > Thus spake Arevalo, Carlos ([EMAIL PROTECTED]): > > Hi Brian, > > > > Well, I made a copy of my dhcpd.conf file, executed the mkdhcpserver and enabled > > tmpfs staging. > > > > Populated ip addresses for the test client I am using, and try an > > install unsuccessfully. > > > > I ran df on the client once it failed and gave me the following > > information: > > > > df > > Filesystem 1k-Blocks Use Available Used Mounted On > > tmpfs 452624 452624 8 100% / > > > > I am guessing that my ramdisk size is too small. So, I edited: > > /tftpboot/pxelinux.cfg/(imagename=compute) and the ramdisk does > > not have a ramdisk_size option defined. > > > > FILE: compute: > > ----------------------------------------------------- > > [EMAIL PROTECTED] pxelinux.cfg]# more compute > > DEFAULT boel > > > > LABEL boel > > KERNEL kernel > > APPEND vga=extended load_ramdisk=1 prompt_ramdisk=0 > > initrd=initrd.img root=/dev/ ram rw ramdisk_blocksize=4096 DISPLAY > > message.txt PROMPT 1 TIMEOUT 50 > > --------------------------------------------------- > > > > How do I increase the size of ramdisk? Or do I leave this alone? > > > > The install fail after the ramdisk got full. Any suggestions? > > > > > > > > > > > > Carlos Ar?valo > > hp > > HPTC > > Phone: 603-884-3390 > > Fax: 603-884-5191 > > E-mail: [EMAIL PROTECTED] > > > > > > > > -----Original Message----- > > From: Brian Elliott Finley [mailto:[EMAIL PROTECTED] > > Sent: Tuesday, July 22, 2003 4:12 PM > > To: Arevalo, Carlos > > Subject: Re: [Sisuite-devel] Multicasting Errors > > > > > > # > > # option-144 tells your autoinstallclient to spool the image into a > > # tmpfs > > # prior to laying it down on disk. It is not certain that this is > > # always > > # a good thing to do. And if you're feeling gutsy and want to try it, > > # be > > # sure that your (memory + swap) is at least twice the size of your > > # image > > # or your image transfer will fail when the tmpfs filesystem gets > > # full!!! > > # If unsure, say "no". > > # > > option option-144 "n"; > > > > > > Thus spake Arevalo, Carlos ([EMAIL PROTECTED]): > > > In regards of this issue, the following is my hardware > > > configuration: > > > > > > Server: DL360 - G3 (2 GB memory), Network Driver: bmc5700 > > > > > > Client: DL360 - G3 (2 GB memory), Network adapter: (embedded) > > > Compaq NC7781 PCI-X 10/100/1000 Gigabit NICs with WOL (Wake on LAN) > > > & PXE support > > > > > > Network Switch: Procurve 2650 (10/100) > > > > > > > > > I have tried setting FULL DUPLEX (on & off) with the same result. > > > I > > > keep getting "pipeline full error messages" on the client. > > > > > > Tried MAX_BITRATE = 20M and it goes through 95% of the download > > > and > > > then it fails/retries. > > > > > > > > > Could it be the network driver on the client side? Or some other > > > issue at the client side? > > > > > > I have updated SI from 3.1.2 to 3.1.4 with the same issue. > > > > > > > > > > > > Carlos Arevalo > > > hp > > > HPTC > > > E-mail: [EMAIL PROTECTED] > > > > > > > > > > > > -----Original Message----- > > > From: Brian Elliott Finley [mailto:[EMAIL PROTECTED] > > > Sent: Thursday, July 10, 2003 6:46 PM > > > To: Arevalo, Carlos > > > Cc: [EMAIL PROTECTED] > > > Subject: Re: [Sisuite-devel] Multicasting Errors > > > > > > > > > Thus spake Arevalo, Carlos ([EMAIL PROTECTED]): > > > > Hello, > > > > > > > > I downloaded SI 3.1.2 and the newly released Flamethrower RPMs > > > > to setup systemimager on a DL360g3 system. > > > > > > > > I got the server installed and running and I could install a > > > > client without multicasting (flamethower) running. However, with > > > > multicasting > > > > > > > enabled, I am getting too many retransmit errors when the > > > > client is trying to copy its main image from the image server, > > > > after the specified number of retries. > > > > > > > > Looking at the /tmp/udp-receiver.compC1 (compC1 = > > > > imageName)....... Error file, I see pipeline full! Errors. > > > > > > > > ... > > > > <<multicastError.jpg>> > > > > The attached snapshot shows the Pipeline full error message. > > > > > > > > My flamethrower.conf file has the FEC (Forward Error Correction) > > > > option set to 8*8/128 & MAX_BITRATE = 40M > > > > > > > > Has anyone seen these errors given by flamethrower? What > > > > parameters should I try to change so that the pipeline doesn't > > > > get full? > > > > > > If you are getting "pipeline full" errors, that is an indication > > > that the client is failing to write the data as fast as it > > > receives it. > > > > > > MAX_BITRATE is a good option to tweak. You might try setting it > > > really low -- seems like 20M works for me on a notebook w/IDE disk > > > acting as server and IDE client. That's extremely conservative, but > > > you can try something really small, and work your way up. > > > > > > What network equipment (NICs, Switch(es), etc.) are you using? > > > > > > -Brian > > > > > > > > > > I am using a DL360 g3 client, running Reh Hat 7.3 SystemImager > > > > 3.1.2 with the latest Flamethrower RPM. > > > > > > > > Thanks in advance for the help! > > > > > > > > Carlos Ar?valo > > > > hp > > > > HPTC > > > > E-mail: [EMAIL PROTECTED] > > > > > > > > > > > > > > > > > > > > -- > > > --------------------------------------------------------- > > > Brian Elliott Finley Phone: 630.803.8183 > > > GPG: 3FF8 D096 0E0C D3F3 29B7 6518 D20B 1931 10F8 EE52 > > > --------------------------------------------------------- > > > > -- > > --------------------------------------------------------- > > Brian Elliott Finley Phone: 630.803.8183 > > GPG: 3FF8 D096 0E0C D3F3 29B7 6518 D20B 1931 10F8 EE52 > > --------------------------------------------------------- > > -- > --------------------------------------------------------- > Brian Elliott Finley Phone: 630.803.8183 > GPG: 3FF8 D096 0E0C D3F3 29B7 6518 D20B 1931 10F8 EE52 > --------------------------------------------------------- -- --------------------------------------------------------- Brian Elliott Finley Phone: 630.803.8183 GPG: 3FF8 D096 0E0C D3F3 29B7 6518 D20B 1931 10F8 EE52 --------------------------------------------------------- ------------------------------------------------------- This SF.Net email sponsored by: Free pre-built ASP.NET sites including Data Reports, E-commerce, Portals, and Forums are available now. Download today and enter to win an XBOX or Visual Studio .NET. http://aspnet.click-url.com/go/psa00100003ave/direct;at.aspnet_072303_01/01 _______________________________________________ Sisuite-users mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/sisuite-users
