From: Charlie.Lai@eng.sun.com (Charlie Lai)
Message-Id: <199903152352.PAA08005@angeles.eng.sun.com>
Subject: Re: Question Regarding SecureRandom
To: csumner@softwarebay.com (Craig Sumner)
Date: Mon, 15 Mar 1999 15:52:31 -0800 (PST)
In-Reply-To: <36E7EC5D.1CBD8F42@softwarebay.com> from "Craig Sumner" at Mar 11, 99 10:16:29 am
> I have an application that creates a secure connection between two VMs.
> On windows NT I can run both VM's on the same box and bring the
> connection up in a reasonable amount of time. If I boot Windows 95 on
> the same machine and then attempt to create a secure connection between
> two VMs running under Win95, the VMs appear to almost deadlock while
> constructing SecureRandom(). When SecureRandom eventually returns, one
> VM appears to be locked. Is there a known problem with SecureRandom on
> Win95 platforms or constructing SecureRandom from two VMs at the same
> time? I've tested the same code on NT, OS/2, and AIX and it works on
> each of these platforms.
are you using JDK1.1.7?
in 1.1.7, we improved the Seed Generator for the SecureRandom implementation.
the improved Seed Generator gives us better random numbers,
and performance is not a problem on Solaris or WinNT.
however, because Win95 does a very poor job managing multiple threads,
and this new implementation spawns a number of new threads,
performance on Win95 is bad.
we have a fix that will go back into 1.1.8.
unfortunately even with this fix, win95 performance
is still about twice as slow as that on NT (on the same physical machine).
however, this is FAR better than 1.1.7, where it seems to hang forever.
if you move to Java 2, the same issue exists.
performance using NT/Solaris is fine.
Win95 is about twice as slow as NT
(but FAR better than 1.1.7).
thanks,
charlie