Hi all, I need some suggestions or hints which syslinux version I shall use. The main focus will be on PXE boot. I have to upgrade due to (expected) performance problems (tftp), so I will swich to HTTP boot ! existing system: - syslinux version 4.05 (tftp only) - used modules : pxelinux.0, memdisk, vesamenuc32, patched startrom.0 - I boot .nbi , .iso, .ima and sometimes files (clonezilla etc.) - size of the .nbi and .iso : up to 25 MB - one main menu (1st. level)and up to 5 submenues (2nd level) - 1 GBit backbone , 100Mbit to the clients - 1000 boots a day , can be 200 boots within half an hour - same files and menus exist 1:1 (without nbi ) on bootable USB-sticks and a multiboot CD. - every kind of PC / Laptop / Server Hardware will be processed (except UEFI only assets). - live and stable in production since 2006 on up to 10 locations. The new systen has following requirements: - get rid of .nbi boot (plus startrom.0) - 2x1Gbit (later on 10 Gbit) backbone, 1Gbit for the clients - the "standard" .iso will grow to 60MB or more - switch from TFTP to HTTP (at least for the .isos) - must be live for production at all locations at the 1st of April 2014 - old (?) problems with some Dell assets with Broadcom Nics must be fixed (syslinux version 4.10) - adaptable to usb-boot and multibootCD - UEFI boot (for production usage) is not part of this project Unfortunally i will get next week the first 60MB.iso for testing. I hope all of this info helps you to give me a good suggestion which version I can / shall implement in my lab for doing the upgrade. bye Christoph
On Sat, Aug 17, 2013 at 6:48 AM, Christoph Hanle <christoph.hanle at leinpfad.de> wrote:> Hi all, > I need some suggestions or hints which syslinux version I shall use. > The main focus will be on PXE boot. > I have to upgrade due to (expected) performance problems (tftp), so I > will swich to HTTP boot !Primary performance issue with TFTP is many PXE systems top out around 1 MB/s, even on Gigabit and 1 switch. Add additional switches, especially a layer 3 routing device, and it's even slower (easily 250kB/s or as low as 100kB/s). Can you deal with a client taking 1-4 minutes to just fetch files? If not, you will have a performance issue.> existing system: > - syslinux version 4.05 (tftp only) > - used modules : pxelinux.0, memdisk, vesamenuc32, patched startrom.0Where does startrom.0 come from? Some Windows file?> - I boot .nbi , .iso, .ima and sometimes files (clonezilla etc.) > - size of the .nbi and .iso : up to 25 MB > - one main menu (1st. level)and up to 5 submenues (2nd level) > - 1 GBit backbone , 100Mbit to the clients > - 1000 boots a day , can be 200 boots within half an hour > - same files and menus exist 1:1 (without nbi ) on bootable USB-sticks > and a multiboot CD. > - every kind of PC / Laptop / Server Hardware will be processed (except > UEFI only assets). > - live and stable in production since 2006 on up to 10 locations. > > The new systen has following requirements: > - get rid of .nbi boot (plus startrom.0)So planning to direct load of kernel/initrd files or use MEMDISK?> - 2x1Gbit (later on 10 Gbit) backbone, 1Gbit for the clientsFor boot, Gigabit to the client may not make too much difference but I'd get your backbone and boot/image server(s) up to 10Gbit before pushing clients up and carefully consider pushing the backbone to 2x10Gbit in some active config (EtherChannel / PortChannel / LACP / Port Trunk). Remember that depending on your balancing algorithm, you could get 1 client flooding link-A, a second client trying to use link-A and link-B being unused.> - the "standard" .iso will grow to 60MB or moreI don't know any tftpd that can't handle this (albeit slowly).> - switch from TFTP to HTTP (at least for the .isos) > - must be live for production at all locations at the 1st of April 2014That's quite a bit of time away. Any idea how much testing needs to be completed first?> - old (?) problems with some Dell assets with Broadcom Nics must be > fixed (syslinux version 4.10)4.10 was never a full release but 5.1x replaced it. I believe this behavior was first observed by someone else on a 960. I'm trying to work through the issue as I've observed it on a Dell OptiPlex 760 unfortunately my debug build stumbled on another bug and availability varies.> - adaptable to usb-boot and multibootCDGenerally, going from PXE to USB/CD is easier as you generally have to follow a scheme of avoiding prefixing with '/' and carefully consider prefixing with "../" on PXE.> - UEFI boot (for production usage) is not part of this projectWhich means 6.0x isn't a required target.> Unfortunally i will get next week the first 60MB.iso for testing.For a first run at this time, I'd consider gpxelinux.0 from 4.07 or perhaps 5.01/5.02-pre3/5.11-pre9 and keep an eye on bug #19 ( http://bugzilla.syslinux.org/show_bug.cgi?id=19 ) for possibly switching to lpxelinux.0. -- -Gene
THX, helps a lot On 17.08.2013 15:09 Gene Cumm wrote:> Can you deal with a client taking 1-4 minutes to just fetch files? If > not, you will have a performance issue. >Unfortunally not, the performance issue is known )-:> > Where does startrom.0 come from? Some Windows file? >I don't know, it is build by the company where we get the .nbis from> > So planning to direct load of kernel/initrd files or use MEMDISK? >Use of MEMDISK as standard and for mass production. kernel/memdisk will be only used for non critical software, but this can change each day.> >> - the "standard" .iso will grow to 60MB or more > > I don't know any tftpd that can't handle this (albeit slowly). >The performance will be the problem. For now we have for more than 50 boots (25 MB images) at the same time a significant performance problem, if additional assets boot there can be a stop until the first assets have finished their booting, independent of the tftp implementation. So i will get rid of tftp.>> - switch from TFTP to HTTP (at least for the .isos) >> - must be live for production at all locations at the 1st of April 2014 > > That's quite a bit of time away. Any idea how much testing needs to > be completed first?(-: the syslinux-part is less than 1/5th of the changings I have to do; and it is not visible to the end users, they expect a running and fast system.> >> Unfortunally i will get next week the first 60MB.iso for testing. > > For a first run at this time, I'd consider gpxelinux.0 from 4.07 or > perhaps 5.01/5.02-pre3/5.11-pre9 and keep an eye on bug #19 ( > http://bugzilla.syslinux.org/show_bug.cgi?id=19 ) for possibly > switching to lpxelinux.0.Yes, makes sense. I think I will use gxpelinux.0 from 4.07 or 4.05 for "quick and dirty" boot and have the 5.1x with gpxelinux.0 or lpxelinux.0 in the lab and testing phase. bye Christoph>