Re: Netware is no VMS
david20_at_alpha2.mdx.ac.uk
Date: 08/07/03
- Next message: Fred Kleinsorge: "Re: Sun tries to woo AlphaServer users away from HP"
- Previous message: Barry Treahy, Jr.: "Re: Netware is no VMS"
- In reply to: Barry Treahy, Jr.: "Re: Netware is no VMS"
- Next in thread: John Eisenschmidt: "Re: Netware is no VMS"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Date: Thu, 7 Aug 2003 10:03:33 +0000 (UTC)
In article <3F313BD6.1010405@MMaz.com>, "Barry Treahy, Jr." <Treahy@MMaz.com> writes:
>jlsue wrote:
>
>>On Wed, 6 Aug 2003 04:15:10 -0400, "Bill Todd" <billtodd@metrocast.net>
>>wrote:
>>
>>
>>>Er, no, it does not. It shows VMS development through *2005* - and with
>>>precious little detail (not even a specific *version* number) about the
>>>projected 2005 release (e.g., *no* new features defined - just 'performance,
>>>security, storage, standards', which could mean next to nothing).
>>>
>>>For example, the following slides show a single 2005 feature - 'new cluster
>>>interconnect' (and that limited to Itanic platforms - so much for Alpha
>>>equivalence) - plus some 'future investigations' which are explicitly stated
>>>*not* to be commitments (not that 'commitment' carries much weight anyway).
>>>
>>>
>>
>>I know this take us a little off-topic, but....
>>
>>What new features do you believe are needed above what's already in the
>>roadmap? For me, at this point, I don't see any real missing killer
>>feature that's both feasible (i.e., not something like process migration
>>between cluster nodes), and valuable.
>>
>>
>
>A very interesting Linux project, which is underway, is a clustering
>project that creates a unified process space between cluster nodes so
>that true load balancing and failover can occur when a member node
>either slows or dies... There is no question that VMSclusters has lead
>the way in clustering, but it was never completed to what would be a
>logical conclusion; Process failover/recovery and true load-balancing...
>
Process migration when one system is overloaded might be possible and
desirable.
Though you would then have to be watch out for a new type of "Thrashing".
System A in the cluster appears to be overload - pick a process to migrate
(making sure of course it is not one of the processes such as system processes
for that system which HAS to run on that system).
System B now running the migrated process.
System B now appears to be overloaded - pick a process to migrate to system A
etc
No work done since cluster is spending all its time migrating processes.
Doing it for a crashing system is much much more difficult.
How much time do you have to migrate the processes ?
How corrupted had the memory structures become before the crash ?
What happened during the inevitable delay between something going wrong and the
OS noticing that there was such a major problem it's only solution was to crash
?
The only solution would be to checkpoint all the processes at intervals to disk
so that they could restart from the checkpoint. You would also need a
journaling filesystem so you could back out all the changes made by the process
between its last checkpoint and the crash. All this checkpointing would slow
down processing.
Also unless the system reviving the crashed processes knew exactly what caused
the crash it is quite likely you would rerun the reason for the crash on the
node you migrated the processes to.
Personally I doubt its feasible and I've probably missed out other problems.
David Webb
VMS and Unix team leader
CCSS
Middlesex University
>So, though I was not asked the question, yes I believe VMS can always be
>improved and features added, and this is just one central core issue
>that is OS specific...
>
>Barry
>
>
>--
>
>Barry Treahy, Jr E-mail: Treahy@MMaz.com
>Midwest Microwave Phone: 480/314-1320
>Vice President & CIO FAX: 480/661-7028
>
>
>
>
- Next message: Fred Kleinsorge: "Re: Sun tries to woo AlphaServer users away from HP"
- Previous message: Barry Treahy, Jr.: "Re: Netware is no VMS"
- In reply to: Barry Treahy, Jr.: "Re: Netware is no VMS"
- Next in thread: John Eisenschmidt: "Re: Netware is no VMS"
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
Relevant Pages
|
