Speed test
Nenad Batocanin
A while ago I sent the test results of different RDD. I noticed that on some machines the results are _very_ different. So I made new tests on different servers. Win 2003 server and its network is weaker than 2008 server. The test consists of a series of simple operations (see attach). The results are very interesting:
- XP machines operate much faster than Win7, even when Win 7 hardware is more powerful (for example, line 4 and 9). I tried to modify some LAN/OS settings, but the results remained the same.
- Direct LAN access is sometimes _much_ faster than client-server! For example, line 14/19. Interestingly, ADS/Leto results are inferior, regardless of the faster network and hardware.
- Leto/ADS is worse than LAN/HBNetIO in appending records.
- Leto/ADS give similar results in all cases, regardless of the change of environment. HBNetIO results also differ slightly and relatively unresponsive when accelerate network/hardware. It is obvious that the direct LAN approach uses a some type of cache that does not affect Leto / ADS / HBNetIO.
Regards, NB
Domenico D'Oria
many thanks for your information
can someone test this code in linux environment ? ( debian, ecc. ecc ) 64 or 32 bit
regards
Nenad Batocanin
Yes, it would be very interesting.
Regards, NB
--
--
You received this message because you are subscribed to the Google
Groups "Harbour Users" group.
Unsubscribe: harbour-user...@googlegroups.com
Web: http://groups.google.com/group/harbour-users
---
You received this message because you are subscribed to the Google Groups "Harbour Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to harbour-user...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Francesco Perillo
Thank you for the tests. They can show differences between OS but I think they cannot be used to compare RDDs. Test are really to simple!
If I will ever start to use hbnetio, for example, I will move all the heavy computing on dbf to the server. There may be a 10x speed increase!
--
Nenad Batocanin
Hello!
I run this test because we noticed a strange behavior of our program in some places. So I tested the basic operation, intending to find out where it is a critical point. This test give me answer and this will help me in the future development.
On the other hand, I do not know the answer, but I'm very interested in why XP work so much faster than Win7. It would also be very helpful to determine why some networks operate much faster than (similar) others and why ADS/Leto/HBNetIO does not take full advantage of that speed?
Regards, NB
Daniele Campagna
Hello!
I run this test because we noticed a strange behavior of our program in some places. So I tested the basic operation, intending to find out where it is a critical point. This test give me answer and this will help me in the future development.
On the other hand, I do not know the answer, but I'm very interested in why XP work so much faster than Win7. It would also be very helpful to determine why some networks operate much faster than (similar) others and why ADS/Leto/HBNetIO does not take full advantage of that speed?
Regards, NB
I'm puzzled. Win 7 seems a lot faster than XP in my experience, in a case the same PC after the upgrade seems to run faster, but no xbase stuff here. I have also an xBase program on a network environment that runs on a bunch of XP machines and one Win 7/64. This too seems to run faster by every respect, database access included, but it is not a valid test because the machine is much more powerful than other PC's in the network (Linux server, btw).
So your tests are very interesting and I hope they are controlled and formalized so everybody could independently confirm (or invalidate) them.
Any other's experience to share?
Dan
Nenad Batocanin
Hello!
The tests I've done on 5 different servers (so far) and I always got similar results - except for LAN and partly for HBNetIO. Of course, this is a very small sample for a permanent conclusions and maybe I'm something doing wrong. I do not think the problem is in Win 7, more likely is a network protocol that communicates with the server.
Test program is very simple and can easily be started on any computer. I will be happy to help if someone wants to test.
Regards, NB
From: harbou...@googlegroups.com [mailto:harbou...@googlegroups.com] On Behalf Of Daniele Campagna
Sent: Wednesday, May 14, 2014 2:30 AM
To: harbou...@googlegroups.com
Subject: Re: [harbour-users] Speed test
Il 14/05/2014 01:35, Nenad Batocanin ha scritto:
Gale Ford
Keep in mind that a data/index file opened in shared mode on shared drive will lock and unlock index file with every skip and seek. This is in addition to record locks for appends and updates to key fields. So if you have many users skipping, seeking, appending, replacing on the same data file there are many file and record locks, unlocks, and possible lock contentions happening. This is where the client/server benefits come into play.
If you use ADS in proprietary locking mode then there is no need for the user machine to lock and unlock during each append, replace, skip, or seek. With 50+ users you would see a dramatic speed difference.
Nenad Batocanin
Yes, I know. But, how to explain this (SKIP test, 10.000 records):
1. LAN 46, HBNetIO 35, ADS 1.19 sec (Win 7/Win 2003)
2. LAN 1.67, HBNetIO 36, ADS 1.89 sec (Win 7/Win 2008)
3. LAN 1.08, HBNetIO 9.94, ADS 1.19 (XP/Win 2008)
Obviously, system (2) have a faster network. But why HBNetIO does not accelerate on the Win7 as on XP?
Regards, NB
From: harbou...@googlegroups.com [mailto:harbou...@googlegroups.com] On Behalf Of Gale Ford
Sent: Wednesday, May 14, 2014 5:10 PM
To: harbou...@googlegroups.com
Subject: [harbour-users] Re: Speed test
Comparing 1 user Client/Server to 1 user direct access will never be as fast.
--
Gale Ford
On Sunday, May 11, 2014 9:14:09 AM UTC-5, Nenad Batoćanin wrote:
Nenad Batocanin
Good question. I used 5 servers, 2 of them have oplocks off, and for the other I'm not sure.
NB
From: harbou...@googlegroups.com [mailto:harbou...@googlegroups.com] On Behalf Of Gale Ford
Sent: Wednesday, May 14, 2014 8:46 PM
To: harbou...@googlegroups.com
Subject: [harbour-users] Re: Speed test
Have you turned off opportunistic locking under Windows 7, Server 2003, and Server 2008?
--
Francesco Perillo
Good question. I used 5 servers, 2 of them have oplocks off, and for the other I'm not sure.
To me 1.67 and 1.08 seconds are wrong numbers, using the wrong oplock values....
Nenad Batocanin
Hello!
All the tests that I run of course is not exact, simply because they do not have the same starting conditions. For example, one of the workstations is i7/32GB/SSD disk, and XP machine is an old AMD with 2 GB RAM. The machine/network settings are not identical. Unfortunately, I could not affect such things. Therefore, the results should be interpreted carefully. But in all these machines already running our software, and I wanted to know why sometimes run faster or slower. Because this test was useful to me and I hope someone else. BTW, oplocks setting has significance only for direct LAN acces, right?
Regards, NB
From: harbou...@googlegroups.com [mailto:harbou...@googlegroups.com] On Behalf Of Francesco Perillo
Sent: Friday, May 16, 2014 1:31 PM
To: harbou...@googlegroups.com
--
Gale Ford
True client/server does not make operating system lock calls to update indexes or skip and seek operations.
ADS can do both. If you open a DBF using ADS in compatible mode then the locks work similar to direct access. The only difference is ADS server is making the lock calls to the operating system. If ADS opens the DBF in proprietary mode then ADS handles all locks and does not have to make the operating system compatible lock calls. That also means there is no reason to do a file lock during every skip and seek operation. With many simultaneous users, proprietary mode can significantly reduce network traffic and lock contentions.
On Sunday, May 11, 2014 9:14:09 AM UTC-5, Nenad Batoćanin wrote:
elch
Hi,
to Gale: yes, that is maybe three quarter of the truth ...
Nenad, did you maybe ment:
L_A_N == L_ocal A_ccess of N_tx from ram ?? ;-)
Further i would suggest, to exclude from your results such parts as needed for time for creating random numbers/ strings.
This could be easy done with creating a big array with these values, before start measuring 'pure' data access time.
-----
I was working on a network benchmark, based on a multithread streaming socket server.
Similar tools sure exists, but attachment is done with Harbour and mostly C-functions.
THAT is the same kind, HBnetio or LetoDB do their communication: client have a question, server then answers.
The shorter the answer, the lower the network traffic -- because there is a limit of answers per time.
If the answers get longer, other limits are reached (CPU, NIC ..)
But such tests will result into a 'performance indicator',
which depends *only* on CPU and NIC power of the client & server hardware combination,
because it is done without any database access.
Together with this value we then can much better compare different pairs of client & server hardware,
when they transfer data from database. So if a client/ server pair have only half of the performance compared to another pair,
we much better can estimate the database time results.
Still work in progress, but for first test should be good enough ...
BTW, my interests in doing this work had been: how much send- and receive- buffers have influence on results (result: less)
-- or if my hardware is capable to transfer zipped more traffic than not zipped, because to zipp needs real CPUs power.
Some explanation at top in source, or ask ...
regards
Rolf
Nenad Batocanin
Hello Rolf,
I am very aware that my test is very rough and has many flaws. Basically I'm measure time in a race car vs bicycle and trying to figure out why sometimes bicycle is faster ;)
But I solved my own very specific problem: on some computers move to client-server processing does not give good results. For example, a report which is generated for 60 seconds, with Letodb/ADS run 200+ seconds on the same machines, with same conditions! Why? Of course, users do not care whether it comes to software, hardware or something else - he just want reports as quickly as possible. That's why I'm doing these tests in this way.
Your program at first glance seems to me very interesting, I'll try it on different systems.
Regards, NB
From: harbou...@googlegroups.com [mailto:harbou...@googlegroups.com] On Behalf Of elch
Sent: Sunday, May 18, 2014 6:37 PM
To: harbou...@googlegroups.com
Subject: Re: [harbour-users] Re: Speed test
Hi,
--
elch
Yep
Przemek is right:
so instead of forcing a 'class C network broadcast, respect the netmask and calculate for Windows
the correct broadcast address ( for Linux they are fine before ).
Then the answers from servers at different networks won't mix trough, looks 'intentional' ;-)
For my personal needs i still like to leave the loopback etc. aside -- but that would be only two IF's to remove ..
Quite happy that it now works like a charme also for Linux
-- now i can start to throw out of the source the 'hardcoded' IP adresses for e.g. HBnetio,
saves me all the configuring work ....
regards
Rolf
elch
Hi Nenad,
some results from Linux -- as ADS is sure no option for me, and 'Samba' also 'somehow' off the table : only checked HBnetio and LetoDB against.
Attached further your test3.prg, a bit modified in the direction like i earlier suggested ..
.. report run 200+ seconds
Nenad Batocanin
Hello Rolf,
Thanks for results and test modification ;) Linux seems like an excellent choice, I thinking to use it.
Regards, NB
From: harbou...@googlegroups.com [mailto:harbou...@googlegroups.com] On Behalf Of elch
Sent: Tuesday, May 20, 2014 4:08 PM
To: harbou...@googlegroups.com
Subject: Re: [harbour-users] Re: Speed test
Hi Nenad,
--