Re: Performance comparison Alpha ES40 vs Itanium rx3600

Stephen Hoffman wrote:
Syltrem wrote:

That server (the ES40 to be replaced by some Itanium box) hosts 6
Oracle databases of which 1 is heavily accessed, plus all clients
(300) that run different applications (mostly 4GL apps), some of which
use RMS files and others Oracle.

A simple answer here is almost certainly a wrong answer.

Whether a move to OpenVMS I64 and an Integrity server will speed your
operations, or hinder them, depends greatly on various factors and data
not yet in evidence.

At the core, the typical OpenVMS I64 configuration is faster than the
typical OpenVMS Alpha AlphaServer configuration, and the Integrity
rx3600 is a very nice and very fast box. But whether it will be faster
for your specific installation and your specific applications is a far
more difficult question.

Hoff finally starts to get to the core of your question. The current
state of the art Integrity servers are on average just incrementally
faster than the Alpha you describe. The thing is that there is no
simple apples to apples comparison. Different kinds of IO perform very
differently between these architectures. Like many have said, your
mileage may very. I am investigating similar questions to yours for
our site. Our Alpha's are much older, so I know I will get a
performance boost, but it is unclear how much and I do not have time to
test drive at this point.

The key, IMHO, is that you should not count on anything but a small
boost in going from your particular Alpha's to fairly cutting edge
Integrity. It just isn't that much faster. On the other hand, if you
are CPU bound on your individual processes at all, it seems likely that
you would get a pretty good boost. If you are IO bound and you are not
upgrading your IO system, maybe not much.

One thing to keep in mind is that while data takes up the same amount
of memory on these 2 systems, executables are around twice as big.
Since you seem to be some what limited by your current memory, you
should definitely get more memory than the Alpha's have.

Also keep in mind that the main thing you get from more expensive
Integrity models is not speed, but expandability. If you can fit all
of the CPU, memory, and PCI cards into a model, then that server is
probably good enough for you. The parts of your config that you list
look to me like you could get by with a 2620 (2 dual core = 4 CPU).
You would need to judge that based on all factors.

A tool like the very ancient "QUALIFY" might be nice, but it's not an
easy problem to solve.

The key consideration here involves what factor(s) are currently
limiting your performance, and whether (or not) an Integrity server will
resolve these. If the application environment is pounding on a disk
spindle, for example, then even massive upgrades in processor speed
probably won't help aggregate performance.

It also has Advanced Server (Pathworks) with 500 connections and 50
open files, and a dozen Progress databases.

IIRC, the Advanced Server product isn't available on OpenVMS I64. The
target Microsoft Windows SMB/CIFS server environment involves an HP port
of the Samba server and its accouterments, and this Samba SMB/CIFS stuff
is AFAIK in field test.

No Advanced Server on I64. CIFS is not due out until 2007. As of now
a migration tool from Advanced Server to CIFS is not promised in the
initial release. Recreating 100's of shares could be very painful. We
are looking at leaving a single Alpha in our new I64 cluster (planned
for next year) just to help with transitions and contingencies. If we
don't need the Alpha, we will not include it, but until that is proven,
we will continue to plan on bring one forward.

Thomas Wirt
Operations Manager
Kittle's Home Furnishings

And also a few other minor things.

The MONITOR SYSTEM reports
Processes - 450
CPU Busy (234) -- 4 CPUs
Direct I/O Rate (3956)
Buffered I/O Rate (2819)

If you want an answer with a degree of certainty (rather than the
guestimates that I or others can provide here), then I might suggest
that there be a prototype attempted. Use one or more of of your
applications -- some sort of a representative sample -- and load it and
test it on the Test Drive system or equivalent HP prototype environment
to see how fast it goes. I'd expect the server will PROBABLY be faster,
though the actual observed aggregate performance could range from
GLACIAL to GONZO, depending on what performance limits exist within your
current configuration. :-)

The other approach is to bring a local, borrowed or leased test system
on-line, and use that to qualify and to stage the migration. You or
somewhat you designate will be tweaking and/or porting code here after
all, and that's a task best done on a server reserved and dedicated for
testing, for target practice and for porting and development; on a
platform entirely disconnected from a production server and its
environment.

There is minimal pagefaulting but not much memory left (XFC cache 77MB
and Free List 18MB)
Those numbers are quite normal during daytime.

An XFC cache of 77 MB does not look to be a particularly large cache,
and would look like a potential performance constraint. Is this
AlphaServer ES40 system already configured near or at the upper limits
of its physical memory addressing? I've seen single-user OpenVMS Alpha
workstations operating with half-gigabyte XFC caches. Cache hit rates
factor in here -- if your database doesn't use XFC, then having a large
cache doesn't matter. But if there are applications that can or do use
the XFC (read) cache, then having an undersized cache forces disk I/O
for those read; cache misses.

.

Relevant Pages

  • Re: PSQL control center crashes PSQL engine.
    ... usually eek out 1.7 GB of cache, but it really depends on other usage, ... memory on the server was just above 2GB. ... Certain versions of Windows do allow a process to allocate mroe than ... If you truly want to use all the memory in your server and you have ...
    (comp.databases.btrieve)
  • POHMELFS high performance network filesystem release.
    ... I'm please to announce POHMEL high performance network filesystem. ... POHMELFS stands for Parallel Optimized Host Message Exchange Layered File System. ... Local coherent cache for data and metadata. ... Very fast and scalable multithreaded userspace server. ...
    (Linux-Kernel)
  • Re: IA64 Linux VM performance woes
    ... the server with over 8GByte memory. ... cache memory pages. ... Linux would have kernel parameter tuning interface. ...
    (Linux-Kernel)
  • POHMELFS high performance network filesystem. Transactions, failover, performance.
    ... I'm please to announce POHMEL high performance network filesystem. ... POHMELFS stands for Parallel Optimized Host Message Exchange Layered File System. ... This is a high performance network filesystem with local coherent cache of data ... Very fast and scalable multithreaded userspace server. ...
    (Linux-Kernel)
  • Re: OLAP Engine Cache
    ... Thanks for the info but we tried the client cache size already and didn't ... We saw some by changing the server side Memory ...
    (microsoft.public.sqlserver.olap)