1 / 16

Window NT 4 Scaling/Performance Tests

HEP NT Days. Window NT 4 Scaling/Performance Tests. Alberto Di Meglio CERN IT/DIS/NCS. Introduction. Windows NT has usually a bad reputation for scaling to large environments NICE is currently based on NetWare, but situation could/has to change in the future

bracha
Download Presentation

Window NT 4 Scaling/Performance Tests

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. HEP NT Days Window NT 4 Scaling/Performance Tests Alberto Di Meglio CERN IT/DIS/NCS

  2. Introduction • Windows NT has usually a bad reputation for scaling to large environments • NICE is currently based on NetWare, but situation could/has to change in the future • The NICE Web Services needs some answers now HEPNT Days at CERN

  3. Windows NT Load Tests • We have started a series of tests on Windows NT as a NICE application server • The test has involved one server and (in principle) about 3800+ clients running W95 (80%) and WNT (20%) • The tests have been run in two rounds for three weeks each. Only the middle week has been considered for statistics (curve is bell shaped) HEPNT Days at CERN

  4. SRV5NICE SRV1_NICE SRV2_NICE SRV3_NICE SRV4_NICE SRV0_NICE Users Migration • Users are normally distributed on the NICE replicas at logon by looking a list of the available servers • The list has been modified for all NT and W95+Microsoft Networks clients HEPNT Days at CERN

  5. Server Specifications • Digital Server 5000 • 1x Intel Pentium II 300MHz • 256 MB RAM • 2x 9GB/1x 18 GB HD on 2x F/SCSI controllers (no RAID) • 1x DC2104 FastEthernet card • 2 partitions: C: (2GB), D: (16GB) with NTFS • Swap file: 256 MB HEPNT Days at CERN

  6. System Specifications • WNT 4 SP3/4 • Home/Profile server with 3000+ exported shares • Performance/Network/Disk Monitors • Real-Time anti-virus scanning • Backup client with daily backups • IIS4 (running, but not used) • Standard server services (server,browser,RPC,etc.) HEPNT Days at CERN

  7. A Simple Simulation Test • A simple user creation program used to artificially increase sessions • Program was running from 6 clients, going through a loop to login a user onto the server and move a few NB of files • From Wednesday 22:30 to Thursday 24:00 HEPNT Days at CERN

  8. Test Areas and Items • Server (Sessions, Open Files) • Processor (%CPU, Interrupt/sec) • Memory (Available Bytes, Committed Bytes, Pages/sec read) • Logical Disk D: (%Disk Read Time, Average Disk Read Queue Length, Disk Read Bytes/sec) • Network (%Network Utilisation, Bytes Total/sec, Bytes Transmitted/sec) HEPNT Days at CERN

  9. Server • Max sessions • (R) 1050 • (S) 2703 • Avg. sessions • (W) 762 • (D) 592 • Max Open Files • 18039 • Avg. Open Files • (W) 14485 • (D) 10563 HEPNT Days at CERN

  10. Memory HEPNT Days at CERN

  11. Processor • Max %CPUs • (R) 46% • (S) 100% • Avg. %CPU • (W) 29% • (D) 19% HEPNT Days at CERN

  12. Logical Disk • Max %Disk Time • (R) 100% • (S) 100% • Avg. %Disk Time • (W) 35% • (D) 21% HEPNT Days at CERN

  13. Network • Max %Utilisation • (R) 9.1% • (S) 8.5% • Avg. %Utilisation • (W) 6% • (D) 4% • Max Total Byte/s • (W) 1.2 MB/s • Avg. Total Byte/s • (W) 0.25 MB/s HEPNT Days at CERN

  14. Comparisons: NT/NetWare • Sessions: 450-550 • Memory: varies with disk size and VLMs, requirements can be high • %CPU time: 15-20% • %Disk Time: 40-60% • %Network utilisation 20-30% HEPNT Days at CERN

  15. Comparisons: NT/UNIX ? • Difficult to compare: the systems work in different ways • Sessions are defined in a different way: • terminals with users using all of the server: a few tens • NFS nodes for file sharing, but not permanent connections: 1000+ registered clients, but how many users? • We are performing tests to evaluate AFS servers as file and application servers, show some performance penalties, but hardware is less performant, still not enough users to have comparable session figures HEPNT Days at CERN

  16. Conclusions • WNT4 has at least the same performance as other NOS and probably better memory management than NetWare • Loading applications from 95/NT client is a bit faster than NetWare (but we’re talking of 2-3 sec) • The right hardware is needed, but scalability not a problem even for large environment (up to 600-800 users/server) HEPNT Days at CERN

More Related