On 1 Oct, Tim Reilly wrote:
> Can anyone confirm this? Honestly, this scares the hell out of me.
Marc's warning is a wise one, but Solaris is not the last or most
perfect word in good deployment options. The Volano benchmarks report
on the performance of Java server environments
(http://www.volano.com/report.html). A look at the page shows a couple
of bits of good news on the Linux side:
1) The IBM JDK1.1.8 shows promising results (but still badly lagging
NT).
2) The TowerJ product is kicking everyone's collective butts in
performance and scalability.
TowerJ isn't a JDK port, it's a commercial (not cheap)
1.1-compliant environment that relies on ahead-of-time compilation and
dynamic loading to do its magic (but can still interpret bytecoded
classes). If you want a serious Java server solution, don't give up on
Linux until you've looked at TowerJ.
Nathan
>
> We're planning on using the IBM 1.1.8 JDK for Linux. In a perfect world,
> we'd be deploying on Solaris, but unfortunately we're a startup company
> and we just can't afford it as of yet.
>
> -Tim
>
> On Thu, 30 Sep 1999, Marc Slemko wrote:
>
>> On Thu, 30 Sep 1999, Tim Reilly wrote:
>>
>> > The company I work for is in the late stages of re-architecting our entire
>> > website using Apache JServ, RedHat Linux 6.0, and MySQL. The code is just
>> > about finished (version 1.0 freezes in the two weeks), and we're trying to
>> > spec the servers we will need for deployment. The problem is that I
>> > really have no idea what class of machine is required.
>>
>> Be very very very cautious about even thinking of deploying any site that
>> has a high load and significant Java code using Linux. This is not a
>> slight to Linux, but simply due to the immaturity and poor performance of
>> JVMs on Linux.
>>
>> The JVMs that are out there either have horrible performance or are
>> unstable under load. The blackdown 1.1 one is pretty stable using green
>> threads, but is horribly slow for IO due to green threads, and IO is about
>> all most server side java code does.
>>
>> The current blackdown 1.2 using native threads (with or without jit) is
>> quite unstable under high load. Some bits of code will crash it fairly
>> reliably, other times it just hangs or SEGVs under heavy load.
>>
>> IBM's JVM is so-so in terms of performance, but may have problems under
>> load due to the immaturity of Linux threads. Until recently, there were
>> also problems that made it not work right with jserv due to it improperly
>> reporting that there was no more data to be read on a socket instead of
>> blocking.
>>
>> In my experience, the price/performance for a sparc box running Solaris is
>> actually _better_ than that on Linux (even if the box is 5x as expensive),
>> and a whole lot more reliable, since the JVMs are so much better.
>>
>> You also need to be very careful about what queries you give to mysql. It
>> does not (at least did not) handle concurrent queries; ie. it finishes one
>> before starting the next, so a single expensive query can kill the whole
>> site for some time, make things backup, etc. As long as everything is an
>> easy select that can be done via indexes, things are reasonable but it
>> requires careful design of accesses and updates.
>>
>> Now, my definition of "high traffic" may or may not be more than your
>> definition. But the above is based on my experience trying to deploy a
>> fairly high traffic site running a JVM on Linux.
>>
>>
>> ----------------------------------------------------------------------
>> To UNSUBSCRIBE, email to [EMAIL PROTECTED]
>> with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
>>
>>
>
>
> ----------------------------------------------------------------------
> To UNSUBSCRIBE, email to [EMAIL PROTECTED]
> with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
----------------------------------------------------------------------
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
