mra@srchtec.uucp (Michael Almond) (11/15/90)
Does anyone know if PC-NFS (Sun Microsystems) runs faster than PC/TCP (FTP Software, Inc.)? We are testing out a copy of PC/TCP and it runs/access files from our Ultrix system slower than our current 3COM server. I thought NFS would be at least as fast as anything written by a company that does DOS software. --- Michael R. Almond (Georgia Tech Alumnus) mra@srchtec.uucp (registered) search technology, inc. emory!stiatl!srchtec!mra Atlanta, Georgia (404) 441-1457 (office) [search]: Systems Engineering Approaches to Research and Development
beame@maccs.dcss.mcmaster.ca (Carl Beame) (11/15/90)
In article <302@srchtec.UUCP> mra@srchtec.uucp (Michael Almond) writes: >Does anyone know if PC-NFS (Sun Microsystems) runs faster than PC/TCP >(FTP Software, Inc.)? We are testing out a copy of PC/TCP and it runs/access >files from our Ultrix system slower than our current 3COM server. I thought >NFS would be at least as fast as anything written by a company that does >DOS software. > > >--- >Michael R. Almond (Georgia Tech Alumnus) mra@srchtec.uucp (registered) >search technology, inc. emory!stiatl!srchtec!mra >Atlanta, Georgia (404) 441-1457 (office) >[search]: Systems Engineering Approaches to Research and Development There is a reason why NFS is slower then 3Com. Basically 3Com protocols and structures are designed sepcifically for the DOS environment. As an example under 3Com a read request is made for a large number of bytes, the 3Com server sends many packets to the client at a speed which the remote client can receive all of them. Under NFS, the servers sends data as fast as possible and the dumb ethernet cards of the PCs can't receive the packets fast enough and lose some, causing ALL of the packets to be retransmitted. Also the maximum read size is 8K. One reason for the difference in speed is the amount of data that can be read or written at one time and wether or not the data is checksummed. All PC based NFS packages checksum the incoming data (I hope!). PC-NFS uses 8K unchecksummed writes and 1K reads InterDrive uses 1K checksummed writes and 1K reads BWNFS uses 8k checksummed writes and up to 8K reads - Carl Beame Beame@McMaster.CA
jbvb@FTP.COM ("James B. Van Bokkelen") (11/16/90)
Does anyone know if PC-NFS (Sun Microsystems) runs faster than PC/TCP
(FTP Software, Inc.)?
I assume you're interested in speed of NFS accesses. At the moment, my
feeling is that BW-NFS is the fastest by a small margin, with PC-NFS and
Interdrive being competitive on reads and short (< 1Kb) writes, and PC-NFS
being good deal faster than Interdrive on large writes. I don't consider
it likely that NFS (an open protocol with end-to-end data checksumming and a
general purpose machine being used as a fileserver) will ever be as fast as
the DOS-specific filesharing schemes (lightweight transports, no end-to-end
checksum, fileserver with optimized disk layout).
James B. VanBokkelen 26 Princess St., Wakefield, MA 01880
FTP Software Inc. voice: (617) 246-0900 fax: (617) 246-0901martino@logitek.co.uk (Martin O'Nions) (11/16/90)
mra@srchtec.uucp (Michael Almond) writes: >Does anyone know if PC-NFS (Sun Microsystems) runs faster than PC/TCP >(FTP Software, Inc.)? We are testing out a copy of PC/TCP and it runs/access >files from our Ultrix system slower than our current 3COM server. I thought >NFS would be at least as fast as anything written by a company that does >DOS software. Technology bigots unite! I like IP and NFS too, but I wouldn't make the mistake of thinking that it puts less overhead on the system. By the time you cut down the protocol stack, dedicate the server to LAN operations, and run an over ambitious bootstrap (DOS) on the client instead of an operating system, it's amazing how LAN performance can be tweaked! Even 3Com's old dos-based 3+ Share, with the overhead of XNS managed to deliver reasonable performance, whilst the all-singing, all-dancing (we distribute it, we're bound to make this comment) 3+ Open is PDQ in most circumstances. I wouldn't want to use OS/2 on my workstation, but I have no objections to seeing it on a file server. Just to clarify the issue, I divide my (network) time between IP, IPX, XNS, and occasionallly NETBEUI/DLC and NBP as well, so I'm honestly not protocol biased...... P.S. Has anybody out there got KA9Q to broadcast RIP packets properly when routing? -- +------------------------------------------------+-----------------------+ | I've got a little black book with my poems in, | Martin O'Nions | | I've got a bag with a toothbrush and a comb in,| Logitek Group Support | | When I'm a good dog they sometimes throw me a | martino@logitek.co.uk | | bone in (Roger Waters - The Wall) | 0257 426 644 | +------------------------------------------------+-----------------------+
karl@naitc.naitc.com (Karl Denninger) (11/21/90)
In article <2742AF8F.813@maccs.dcss.mcmaster.ca> beame@maccs.dcss.mcmaster.ca (Carl Beame) writes: >In article <302@srchtec.UUCP> mra@srchtec.uucp (Michael Almond) writes: >>Does anyone know if PC-NFS (Sun Microsystems) runs faster than PC/TCP >>(FTP Software, Inc.)? We are testing out a copy of PC/TCP and it runs/access >>files from our Ultrix system slower than our current 3COM server. I thought >>NFS would be at least as fast as anything written by a company that does >>DOS software. >> >>--- >>Michael R. Almond (Georgia Tech Alumnus) mra@srchtec.uucp (registered) >>search technology, inc. emory!stiatl!srchtec!mra >>Atlanta, Georgia (404) 441-1457 (office) >>[search]: Systems Engineering Approaches to Research and Development > >There is a reason why NFS is slower then 3Com. Basically 3Com >protocols and structures are designed sepcifically for the DOS environment. >As an example under 3Com a read request is made for a large number of bytes, >the 3Com server sends many packets to the client at a speed which the remote >client can receive all of them. Under NFS, the servers sends data as fast as >possible and the dumb ethernet cards of the PCs can't receive the packets >fast enough and lose some, causing ALL of the packets to be retransmitted. >Also the maximum read size is 8K. > >One reason for the difference in speed is the amount of data that can be >read or written at one time and wether or not the data is checksummed. >All PC based NFS packages checksum the incoming data (I hope!). > > PC-NFS uses 8K unchecksummed writes and 1K reads > InterDrive uses 1K checksummed writes and 1K reads > BWNFS uses 8k checksummed writes and up to 8K reads > >- Carl Beame > Beame@McMaster.CA There are a few other considerations: 1) 3COM and Novell will be faster on a one-server one-client basis. HOWEVER, both of these will slow down as you add more users to the network. We can easily support 70 users on one Sparcstation 1+ without trouble, and the load appears to be nearly constant (ie: from the workstation end, there is no degradation of response noted as load increases). That you'll never accomplish with Novell or 3COM. 2) NFS has essentially no overhead until you actually DO file I/O. Both of those other networks can't claim this. This gets to be a real problem as you add users to a 3COM or Novell network. 3) You can throw really BIG iron at a NFS server, with multiple Ethernet segments or even (gasp) FDDI if you want. That's just not possible with either 3COM or Novell. When you get LOTS of people on a big network, this makes an enormous difference. 4) There's an economy issue. A MIPS Magnum costs some $8k. It has something in excess of 20 MIPS of performance, 1.5MB/sec disk transfer rates, and ethernet transfer rates exceeding 600KB/sec. You can't buy that in a PC-based file server for ANY amount of money; you can't even come close for $8,000. And it comes with the server side of the network software, a real operating system, X11, and enough internetworking to allow you to connect it to everything else in your enterprise! (As an aside, I wish someone in the PC Marketplace would come up with a REAL Ethernet interface. PS/2's are worse than ISA machines in this regard, which is quite a surprise.) Both 3COM and Novell were designed for relatively small networks. Witness Novell's failure to provide internal checksums on their IPX protocol; this makes bridging real fun, especially over long distances..... NFS, on the other hand, runs very well even over large networks. It's response does degrade, sure, but it still is quite servicable. And the Ethernet capacity of most decent workstations or servers far exceed that which a PC can grok. PC-based networks (ie: 3COM and Novell) are handicapped by their lack of ability to push the bits down the wire at anything approaching the 10Mb/sec that Ethernet is capable of. Do realize, folks, that 1MB (megabyte now) per second is all Ethernet can do. That's not that fast in today's world of 2MB/sec SCSI disk transfer rates. A PC is lucky to get 150KB/sec out of an Ethernet adapter, regardless of which LAN software you layer on top of it. We have workstations here which can do 500KB/sec, and a few machines which can hit over 600 consistantly, given that the LAN load doesn't cause too many collisions. The extra wire throughput is of great help when you try to server a lot of users off one server. We find here that the NFS lan is FASTER than 3COM. We have both up and running. BWNFS also leaves you with 570K of free RAM, while 3COM wants to eat up about 100K more; this is a BIG difference when you run hungry applications on DOS machines. Be careful of generalizing one-to-one comparisons to many-to-many. We split NFS file service seamlessly over some 30 machines here, and feed MACs, PCs and Unix machines of all descriptions. Try THAT with 3com; you'll need to buy 30 copies of 3+Open, you lose the transparency, and can't do the Unix side of it at all! -- Karl Denninger AC Nielsen kdenning@ksun.naitc.com (708) 317-3285 Disclaimer: Contents represent opinions of the author; I do not speak for AC Nielsen on Usenet.