6 ms·
The flip side, is for the cost of 1 mainframe you can often buy 500 linux PCs.. which end up giving you 100x the CPU and 100x the IO of the mainframe (with a bi
by brianwawok 10y ago
The flip side, is for the cost of 1 mainframe you can often buy 500 linux PCs.. which end up giving you 100x the CPU and 100x the IO of the mainframe (with a bit of other challenges like orchestration and such).
- dekhn 10y agosure... but I'd rather have 1 mainframe than 500 PCs. They typically have higher utilization, the problems don't need to be parallelized, you don't have to think about your network. I still use PCs for everything I do, haven't yet found a need for a mainframe personally.
- kpil 10y agoI have 1 mainframe, and which I had 500 PCs... It's solid engineering, but it's also a lot of branding. If you have a "golden egg" application that have a lot of internal state so you can't run it on 10 servers, then a mainframe might be a good solution to keep it running. I think the most common reason to have a mainframe is that you happen to have too many managers that used to be mainframe developers.
- kpil 10y agoAnd besides, I think you get at least 500x CPU. It's hard to beat Intel's CPUs.
- dekhn 10y agomainframe CPUs are no joke
- kpil 10y agoPer core? Numbers please. The zLinux CPUs are slower than my laptop.
- dekhn 10y agoThe processing unit in a z13 is a 5GHz, with up to eight cores per unit. There is an enormous cache hierarchy (much larger than largest Xeons) often with dedicated per-core cache resources. Are you working from the same data as me? Also, those CPUs, since they are hooked up to IO processors, tend to have higher utilization (less time waiting for data).
- PopsiclePete 10y ago500 linux PCs...Can you imagine a Beowulf cluster of these??
- brianwawok 10y agoOh god that went on for years...
- kjs3 10y agoIt's dealing with that "bit of other challenges" that's worth the extra money. The mainframe folks have been doing large scale orchestration over massive numbers of VMs since the 60s.
- tracker1 10y agoDepends on the data you're working with... transactions across many servers is a hard problem, with huge overhead.