Compaq Proliant DL580 Server - PCI Hotplug does it Actually Work? 

great demo there victor, Windows 2003 server, was stated once the system was up and running, you could actually swop out the display card for a network card. You would log into it using the Remote desktop / terminal service that is enabled by default for administrators, only two remote connections anymore you would need a licence.

I've swapped a NIC or two, but like you said, there's a limited number of things that support this trick on a PC. (esp. these old Compaq's) However, I do recall a story from Ziff-Davis (benchmark lab) where a CPU card caught fire (back in the PPro days.) Without skipping a beat, that card was shutdown, pulled, and new card put in. (that CPU card was fine after the heatsink goo was replaced - that's what burned.) I _watched_ a DEC demo of an Alpha Server where everything short of the backplane could be hotswapped. And they did swap out _everything._

Owned and operated a LOT of Compaq servers. Never once did any hotplug. It struck me as an absolutely terrible idea.

Never in production but I have accidentally knocked out a PCI network card from a powered on desktop PC before and plugged it back in and it worked!

That piggybank easter egg is on the cache module. Cash/cache... it's a homonym.

My first engineering Job was deploying 6 of these connected over fibre to a Compaq SAN, and tape library. Nothing like having to assemble 6 of these with quad Xeon's and add in cards, etc.

We all love Chipmunk Victor!

glad you like the cards :)

I did love working on those DL580 (and it's predecessor, the 6400R), lots of space, but the storage is such a tiny part of the whole chassis. I only really ever hot plugged a NIC once, for anything else you needed to shut down anyway. Hot plugging PSUs was nice and easy though too!

hot plugin saved the day!

PCI Hotplug in the PROD environment in the bank's infrastructure, I would like to mention the good times

Awesome video dude!

I have IBM server with PCI-X hotplug. Never used them. I've seen that it is also possible with PCI-E server hardware. Also I've maxed memory on same DDR1 server to 12GB with 32 bit CPU and OS :) Would be nice to see 16GB on SD-RAM server. Recently i've bought about 30 sticks of 256MB ECC (unregistered) SD-RAM for like 160 USD, double sided ones - compatible with 440bx ;) 16GB of SD-RAM for 320 USD is not that bad i guess for people like us :D 4 sticks of 1GB SD-RAMs 1 year ago cost me like 50 USD, that was a good deal ;)

using the pre-notify hot-plug slot controller (push the button, wait for ack, replace card, push button) is a best-case scenario. NVMe drives with I/O errors can drop off the PCIe bus or timeout, so we get surprise hot-remove and -add events necessitating downstream port containment (DPC) and some fairly involved flows between UEFI/ACPI and OS. A SCSI array with striped mirrors (raid 10) between two controllers (hot-swap one of the controllers) would be a fun use case to really stress-test the hot-swap capabilities and storage subsystem. Ideally storage subsystem would automatically identify and re-sync drives when the second controller was brought back online.

I'm a DCT. I only ever did hot-plug in a lab environment. Production admins seemed utterly unaware that hot-plug was even a thing, and every hardware issue was a cold swap because of it. They were supposed to have failover servers for that reason.

I plugged a PCI SCSI card into my Dell workstation the other week. I thought I'd powered it down but apparently I didn't. Anyway, device manager detected it & it worked fine without rebooting.

Could swear blind I commented on this... Hot plug. Designed as a resiliency function, to minimise downtime, not a pure redundancy function. The Pentium Pro-based *Compaq* ProLiant range from 1997/98 incorporated hot-plug PCI cards, disk drives (SCSI), RAM, and CPU boards. The concept was downtime was minimised, or even avoided, for certain hardware failure scenarios. And given how long some of these beasts could take to boot whilst awaiting every drive in an array to spin up, the cost savings of avoiding downtime far outweighed the cost of technology implementation. I managed the server estate for the UK's largest single contributor to GDP n 1997/98, and every single machine had a spare NIC, CPU card, SMART2 and DIMMS assigned it. Yes, you could hot-swap a SMART2-DH - we did so. The array configuration could be stored on the array or the card, which assisted greatly when migrating, and the BBU really came into play in a failure scenario when there had been a lot of writes (that failure could be software related, ie ABEND or a BSOD) as the data was intact. But, one could swap out a SMART2, ideally swap over the BBU into the new card, put it back in, and off it trundled. At least it did under Netware, cant speak about the abomination that was NT4 ;) SMART2-DH with 16MB Memory and BBU Allowed for 16 devices (including the card itself). The card was 40mb/sec Ultra-Wide SCSI, 40mb bandwidth The two interfaces were effectively the middle of the chain, one could: - connect up to15 devices externally - connect up to 15 devices internally - connect up to 15 devices internally or externally You just had to remember which devices (or the card) to terminate (or not) The 4200 was a quad channel Ultra3 (compaq termed it wide-ultra2) card, allowing 80mb/sec per channel Shipped with a 64mb BBU cache, but 8mb was reserved for buffer and device RAM Ignore Compaq's use of "wide" by definition all SCSI after wide-ultra Ultra2 was 16bit (or 32bit) and therefore 'wide' by default. Again, not all the interfaces could be used: one internal and one external interface, or both external interfaces, per channel. Think of it as two SMART2's on a single board.

Great video✌

Hotplugging never really took off with x86 hardware.. other real server architectures (Sun SPARC, IBM Power..) supported hotplug for almost all hardware, especially CPU and memory boards, which allowed for live upgrades and field replacement of defective parts. With x86 and especially Windows, you were always expected to shut down production before doing any maintenance on your machines. Also you can of course hotplug SCSI cards, just not the one with your running OS on it.

That is a pretty cool server but it is ungodly loud. There is PCI-E hotplug too. The SCSI adapters are monsters. I ran SCSI back in the day. Expensive, but freed you up frim IDE limitations.

PCI hotplug probably doesn't need much special software support. CardBus expansion slots on laptops are basically hotplug PCI devices which work well on the regular consumer OS. On today's servers PCIe hotplug is probably used with NVMe drives.

its just a little click for a man, but such impressive for mankind

The software side of this technology still lives on, you can "hot-plug " virtual devices in to a VM

Tried it once when we decommissioned our Exchange 5.5 server - resulted in a Blue Screen once I inserted the FDDI network card back in

Leuke ouwe meuk - hoeveel db is dat ? ik vond voor een leuke $ een SuperMicro X9SRI-F

👍

Where are my 4*900 mhz slot 2 xeons??

open it up already victor bart

vmware and vmotion ultimately replaced the need for such features