> On 3 Aug 2023, at 2:26 am, P H > <[email protected]> wrote: > > The numbers quoted by Tom: > > So I pointed out there's only 12 I/O drawers max on a z16 which is 12 x > 16 = 192 slots or 384 ports max. He replied, but didn't seem to fully > accept that answer. > > are 100% correct. These numbers are the MAXIMUM. Depending on the > configuration, these could be a lot less e.g. the number of coupling links > could reduce the numbers. If z16 is ordered with BPA power supplies, the MAX > I/O drawers go down from 12 to 10. > > I have already mentioned things like cache, memory, I/O Subsystem, on chip > data compression/Crypto (z has been a leader for this)/Sort/AI capabilities.
Maybe for crypto but certainly not for AI. My iPhone has a more sophisticated AI engine than the z16. Other platforms have integrated AI engines, AMD ZenDNN, Intel oneDNN etc. Both ship with open source libraries and toolkits sadly lacking for z/OS. I noticed that IBM have shipped patched Python packages for TensorFlow and SnapML that exploit Telum for Linux on Z. I suppose like everything, we’ll have to wait a while for z/OS. Java 11 still does not utilise zEDC compression on z/OS. Talking about compression and crypto, Intel have hardware accelerators as part of QAT, either PCIe cards or on-die. You could argue that the compression tech is better than zEDC as it supports more formats then just gzip. https://www.intel.com/content/www/us/en/architecture-and-technology/intel-quick-assist-technology-overview.html > > Talking about the I/O Subsystem, this is a key strength when it comes to > handling large number of I/Os. Unlike x86, the I/O Subsystem handles this > very well and lets the CP get on with what it's mean to do. What no one has > mentioned is the 'processing' power of z. In addition to the main CPs (up to > 200 for z16 Models A01 and L01), the I/O Subsystem has up eo 192 POWER > processors. These are in a N+1 config making a total of 384 in he sub-system > alone. > > Impressive numbers. What do all these prove? Taken out of context, these are > meaningless. As I stated previously, one has to consisder the whole system. > This is where z has strengths. It has a 'balanced system design'. This > morning I decided to do a full virus scan on my 2 year old latop with an > Intel i5 chip. While the scan was running, I couldn't even open a 10 MB > Powerpoint presentation 🙁 (before the smartones give me their 2 cents worth, > I know I could have run the scan as a background task). > Get yourself a better machine. My Mac runs clusters of Linux systems on Kubernetes running stacks like Kafka, ELK at a full pelt without breaking a sweat and I can watch YouTube in 4K at the same time. For a more apples to apples comparison to x86 it would be more interesting to compare a z box to an HP Superdome kitted out with all the fruit. There are only three large systems remaining since Oracle killed off the SPARCs. Z, POWER and the super domes. The server market is dominated by single socket rack servers running distributed systems. > Talking about numbers, the Airbus A380 plane has been designed to have up to > 840 passengers. Are there any airlines with A380s which carry such numbers! > > Horses for courses!! > > ________________________________ > From: IBM Mainframe Discussion List <[email protected]> on behalf of > Tom Brennan <[email protected]> > Sent: 02 August 2023 17:34 > To: [email protected] <[email protected]> > Subject: Re: Channelized I/O WAS: Mainframe Makers.... WAS: Ars Technica: The > IBM mainframe: How it runs and why it survives > >> I’ve missed this thread. > > He first said 1536 ports (not slots, not lanes) on a full z16. I asked > where he got that number. Response was there are 12 fanout slots on a > CEC drawer (true), so with 4 CEC drawers that's 48 fanout slots (true) > which means the 4 CEC drawers could address 48 I/O drawers with 16 cards > each and 2 ports per card = 1536 ports. > > So I pointed out there's only 12 I/O drawers max on a z16 which is 12 x > 16 = 192 slots or 384 ports max. He replied, but didn't seem to fully > accept that answer. > > Later he said there are 1600 slots (not ports, not lanes) on a z16 so I > asked where he got that new number. He said he meant 1536 slots (not > ports, not lanes) so the number doubled from last time. I replied same > as I did previously. > > Below, he said 1536 slots again. 1536 cards on a single z16 could be > over 3000 cables! I've had to untangle some 150+ cable rats nests, but > for that one I'd just say, Naw... I'm going home :) > > On 8/2/2023 1:53 AM, David Crayford wrote: >>> On 2 Aug 2023, at 12:15 pm, Tom Brennan <[email protected]> wrote: >>> >>>> The IBM z16 can have up to 1,536 PCIe+ slots >>> >>> I'm gonna quit explaining this and just say, "WRONG" every time you say >>> this as if it's a fact :) >> >> I’ve missed this thread. By 1,536 PCIe slots, that’s slots not lanes right? >> Even if it were lanes that would be a ludicrous suggestions! That’s so far >> fetched it’s laughable. The Redbook [1] is quite clear about I/O >> configurations. What I find interesting is that the z16 seems to use PCIe >> gen 3 and not gen 4 which doubles the transfer rate per lane. There must be >> a good technical reason for this. >> >> [1] https://www.redbooks.ibm.com/redbooks/pdfs/sg248951.pdf >> >>> >>> On 8/1/2023 8:01 PM, Jon Perryman wrote: >>>>> On Tuesday, August 1, 2023 at 05:20:33 PM PDT, David Crayford >>>>> <[email protected]> wrote: >>>>> What’s the difference between between channelized I/O and a rack of >>>>> x86 servers connected to a SAN using fibre channel driven by high speed >>>>> HBAs? >>>> PCIe was created specifically for PCs and IBM z16 chose to use that as >>>> their only channel technology. Channelized I/O for PC has been available >>>> for several decades and is not limited to PCIe. The IBM z16 can have up to >>>> 1,536 PCIe+ slots. >>>> As for x86 fiber channel connection to a PC, PCIe is only one possibility. >>>> ---------------------------------------------------------------------- >>>> For IBM-MAIN subscribe / signoff / archive access instructions, >>>> send email to [email protected] with the message: INFO IBM-MAIN >>> >>> ---------------------------------------------------------------------- >>> For IBM-MAIN subscribe / signoff / archive access instructions, >>> send email to [email protected] with the message: INFO IBM-MAIN >> >> >> ---------------------------------------------------------------------- >> For IBM-MAIN subscribe / signoff / archive access instructions, >> send email to [email protected] with the message: INFO IBM-MAIN >> >> > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
