Linux 5.15 Addressing Scalability Issue That Caused Huge IBM Servers 30+ Minutes To Boot

[F.Ultra]

Yet another problem helped with semaphores. Something the C standard idiots eschewed and only gave us awful condition variables which are not the same thing as semaphores. I guess everybody is still using pthreads anyway because the standard is so lacking in features. I'm not sure why they bothered to put anything thread related in there.

The kernel does not use pthreads nor the C standard thread.h functions but their own locking mechanisms. In this case the performance gains was from changing the locks from a mutex to a readers/writer-lock so that multiple accesses to kernfs could be done in parallel if they where reads.

[JustRob]

One or two boots per year and your 5 9's are for naught: https://en.wikipedia.org/wiki/High_a...ge_calculation

IBM uses Parallel Sysplex for high availability: https://www.ibm.com/docs/en/zos-basi...rallel-sysplex

[linner]

Try the Tandem/Compaq/HPE "NonStop" servers. The NSK (NonStop Kernel) based servers have been doing 5 nines for a few decades now.

Do you know anything about the software that runs on those NonStop systems? What programming languages are used?

According to Wikipedia, Trafodion uses C++ and Java but I can't imagine most of the NonStop stuff uses those, being as old as the project is. ? Years ago I read articles stating that Linux was then the primary player for stock exchanges. Would that be Linux running on NonStop hardware or is that a whole different thing?

[tildearrow]

First show me another vendor with 5 nines availability.

HPE NonSt- Oh never mind......

[Sonadow]

Do you know anything about the software that runs on those NonStop systems? What programming languages are used?

Who cares about what OS is used as long as it does its job.

Heck, if would be a gratifying slap to Linux's and Unix's faces if those servers used to run Windows HPC.

[linner]

Who cares about what OS is used as long as it does its job.

Heck, if would be a gratifying slap to Linux's and Unix's faces if those servers used to run Windows HPC.

I'm more interested what applications run on it and how they are designed. Not so much the OS. The OS comment was just meant as a side remark. it's good to learn what we can from successful projects and try to do the same or better in the future. Knowledge is power.

[partcyborg]

First show me another vendor with 5 nines availability.

Lol 5 nines is ~5min of downtime per YEAR. Good luck getting that with a 30min boot time

[BillT]

Many years ago when I worked on Tandem gear, the OS was a proprietary system called Guardian. The programming was done using TAL = Tandem Application Language.

[mdedetrich]

Lol 5 nines is ~5min of downtime per YEAR. Good luck getting that with a 30min boot time

Well typically you have redundancy, i.e. multiple servers per "system" or "application" so unless ALL of them are offline at once then this is a non issue.

Clearly a lot of people on this forum have no idea how to design backend systems/mainframes with high SLA. 5 nines is definitely very high but its not impossible like some people claim it to be, at least if you have enough resources to throw at the problem.

[cyberwizzard]

One thing I'm missing in the article: is this a regression? Or was it always as slow as this but recent kernels made it worse? Or was booting slowly the status quo and this is now getting optimized?