I run on zD&T machine on Linux. z/OS starts in about 60 seconds. I'm trying to find out why z/OSMF does not work with Elliptic Curve certificates. With shared cache, and all I/O in ZFS cache - it takes over 2 minutes to start ( and this is the thinnest possible, and no plug ins!). This problem is really tedious to debug.
It would be good if IBM ran for z/OSMF/ZOWE/etc for days, so the JIT was as optimized as possible. Then save the shared classes and ship them as part of the product. Everyone gets a fully optimised system from day 1. I haven't looked into the levels available with shared classes - but having one level for the JVM, and another level for z/OSMF etc might help. Colin On Tue, 14 Jan 2025 at 03:17, Andrew Rowley <[email protected]> wrote: > On 11/01/2025 7:33 pm, Colin Paice wrote: > > The idea says use the existing Java support to harden the shared > > information across IPLs. This means the first start of z/OSMF will be > > slow, as today, the next start will use half the CPU. > > > > Please vote for it, or add comments to it. > > > > I'll look into ZOWE - it doesn't even have shared library support > enabled! > > I looked into the Java shared library support. The thing that deters me > is the warning in the manuals about ahead of time compilation being less > optimized than the JIT compilation so the AOT code may be slower. It's > easy to measure startup time, but not necessarily easy to measure > whether it has an impact on later CPU usage. But if z/OSMF is already > using shared classes it probably makes sense to keep them across an IPL. > > On my system z/OSMF starts in 45 seconds, with 48s zIIP time and 3s CP > time. That's not too bad, but I can see how it's a problem if you have > no zIIP and reduced speed CPs. > > I actually have no idea what z/OSMF could be doing that takes 50s CPU > time... maybe I'm underestimating it's complexity, but I have a > suspicion that there is something dumb happening there. That's > reinforced by type 92 SMF records which suggest that the same doc.zip > files are being read over and over again - thousands of times! > > -- > Andrew Rowley > Black Hill Software > > ---------------------------------------------------------------------- > 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
