Dnes mi v mailu přistála zajímavá diskuse o architektuře WLS domény pro klíčovou aplikaci. Řeší se v ní v podstatě rozhodování mezi dvěma možnostmi: je lepší pár nadupaných serverů, nebo více menších?
Autorem odpovědi je Will Lyons a naprosto s ním souhlasím.
Jsem línej to překládat, takže pastuju originál:
> We are evaluating design of new weblogic domain which we want to be
> created as our servers are reaching EOL. Based on our requirements
> following two approaches were finalized.We want to know from you the
> pros and cons of having these designs.
> 1.)All managed servers are hosted on 3 boxes which have 140GB RAM each
> box(including OS) and it is completely occupied.
> 2.)All managed servers to be distributed across boxes which have 20GB
> at max(including OS).(Effectively in this case in order to make it 140
> we need 7 boxes,multiply it by 3 to get 420 effectively which makes
> the value to 21).
...
My response would be:
- Your question implies that you have a single domain/cluster with at
least 20, or 40, or 60, or 80+ managed servers with XGB of memory per
WLS managed server process. Please confirm/deny and clarify this at a
minimum.
- Since your question implies a cluster/domain with at least 20 managed
servers, this suggests a large and therefore an important
configuration. To obtain a better answer to your question, you should
compare the details of the current configuration with the alternative
projected configurations. Comparing alternative at this level of
detail is more likely to identify potential issues:
- How many domains
- How many clusters/per domain
- How many WLS managed servers per cluster and per domain
- How many apps/WLS managed server
- For apps being deployed in the same managed server, is this a
deliberate choice, and are there specific requirement that they be
deployed in the same server
- How much memory/WLS managed server
- How much memory is available on the physical server for WLS
managed server processes (after use of memory by OS and other software
processes running on the physical server)
- What version of WLS, OS, JVM and are they all supported
- How are you/will you spread these domains/clusters/servers/apps
across the servers in your current and future alternative configurations
- Are you/will you use of multicast vs. unicast messaging for your
clusters
- Are you/will you use the WLS Admin Console for
monitoring/managing domains in production
- Do you have any evidence of any capacity limits or bottlenecks in
your current system or in the projected system(s)
- Having said the above:
- It seems likely that a "21 physical server" configuration would
be sub-optimal (too few WLS managed servers per physical server and/or
too many managed servers per domain, and/or too many physical servers
per domain)
- As a rule of thumb, there will be less network traffic when
running a given WLS configuration on fewer physical servers with greater
processing capacity. So the 21 physical server configuration has more
risk for network bottlenecks.
- You should consider the HA considerations (impact of planned and
unplanned downtime on WLS application availability) based on the number
of physical servers you select. For example, is 3 servers enough? It
may be.
- Your question implicitly assumes that the [(WLS application
processing capacity)/(GB of memory on the physical servers)] is a
constant across all the physical servers you are evaluating. This
assumption is probably invalid or overly simplistic.
Žádné komentáře:
Okomentovat